0 évaluation0% ont trouvé ce document utile (0 vote)

14 vues263 pages765432

© © All Rights Reserved

PDF, TXT ou lisez en ligne sur Scribd

765432

© All Rights Reserved

0 évaluation0% ont trouvé ce document utile (0 vote)

14 vues263 pages765432

© All Rights Reserved

Vous êtes sur la page 1sur 263

préparée dans le cadre d’une cotutelle entre l’Université

Grenoble Alpes et l’Université Fédérale de Santa Catarina

Spécialité : Industrial Engineering

Arrêté ministériel : le 6 janvier 2005 - 7 août 2006

Présentée par

Francielly Hedler Staudt

et codirigée par Gülgün Alpan

dans les Écoles Doctorales: IMEP2 et PPGEP

methodology to determine an inte-

grated performance measurement

devant le jury composé de :

Maître de Conférences, HDR, Université du État de Santa Catarina, Rapporteur

Mme. SAHIN Evren

Professeur, École Centrale de Paris, Rapporteur

M. CURSI José Eduardo de Souza

Professeur, INSA Rouen, Examinateur

Mme. SILVA Vanina M. Durski

Maître de Conférences, Université Fédérale de Santa Catarina, Examinatrice

M. FOLLMANN Neimar

Maître de Conférences, Université Tecnologique du État du Paraná, Examinateur

Mme. DI MASCOLO Maria

Directrice de Recherche, CNRS , Directeur de thèse

M. RODRIGUEZ Carlos M. Taboada

Professeur, Université Fédérale de Santa Catarina, PRESIDENT, Directeur de thèse

Mme. ALPAN Gülgün

Maître de Conférences, HDR, Grenoble INP, Co-Directeur de thèse

Contents

Front page i

Table of Contents v

Acknowledgements ix

List of Acronyms xi

1 Introduction 1

1.1 Context of the study . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2.2 Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

ii CONTENTS

2.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

gration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

3.3.2.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

3.3.3.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

3.3.4.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

3.3.5.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

3.3.6.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

3.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

mance 57

CONTENTS iii

indicators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

4.3 Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

4.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

5 Conceptualization 75

5.1 Introduction - the Standard Warehouse . . . . . . . . . . . . . . 76

Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

5.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

6 Modeling 97

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

6.3.1 The data associations . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

iv CONTENTS

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

7.2 Analysis of Jacobian and Correlation matrix to improve PCA

results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

7.3 Integrated performance model proposition . . . . . . . . . . . . 133

7.4 Scale for the Integrated Indicator . . . . . . . . . . . . . . . . 136

7.6 Model Update . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

8.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150

8.2 Future Research Directions . . . . . . . . . . . . . . . . . . . . . . 153

8.2.1 Short-term Research Directions . . . . . . . . . . . . . . . . . . . . . . 153

Bibliography 166

A Complete Analytical Model of Performance Indicators and

Data 1

A.1 Time indicator model . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

B Data Generation 21

B.1 Receiving data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

CONTENTS v

C.1 The Manual Procedure . . . . . . . . . . . . . . . . . . . . . . . . . 30

E Theoretical Framework of indicator relationships 35

F Results of Dynamic Factor Analysis application 37

G Results of Anderson Darling Test 39

H Optimization model 45

I Mean and standard deviation values of indicators 51

J Optimization results 53

K Abstracts 55

K.1 Contexte du Problème de Recherche . . . . . . . . . . . . . . . . 60

méthodologie . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

logistique de façon agrégée . . . . . . . . . . . . . . . . . . . . . . 64

mance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

miers résultats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

K.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

To my beloved family...

Acknowledgements

Life is made of moments. So, I would like to sincerely thank everybody that shared with

Em primeiro lugar, eu gostaria de agradecer ao meu amado marido Tiago por estar

sempre ao meu lado durante este período. Obrigada por acreditar no meu potencial e

sempre me animar novamente nos momentos em que não acreditava ser possível, por ser

meu incentivador, meu professor, meu conselheiro, meu companheiro, meu amante. Viver

Agradeço ao meus pais, Rolf e Karin, pelo apoio no momento em que decidimos investir

nesta louca jornada. Não foi fácil, mas o suporte, auxílio e paciência de vocês, principal-

mente na fase nal de redação da tese, foram importantíssimos. Amo vocês! Agradeço

ao resto da minha família, minha mana Luanna, Darcísio, Sirlei, Tati, Fabiano por todo o

Agradeço aos amigos Mauricio, Lidiane, Fábio, Simoni e Marconi pelos inúmeros mo-

mentos de alegria que passamos. Com certeza o fardo foi menos pesado com a companhia

de vocês!

doutorado. Muito obrigada por acreditar na ideia do meu trabalho e da cotutela, por

brigar por mim quando foi preciso, por me ensinar que devemos acreditar no potencial

das pessoas.

Vanina, e especialmente ao Professor Jovane Medina por ter aceitado realizar o relatório

da tese.

Je voudrais remercier Maria et Gülgün pour avoir fait l'encadrement rester très proche

même quand on était loin. J'ai appris beaucoup avec vous et j'espère vraiment qu'on puisse

G-SCOP, même quand mon travail n'était qu'une idée. Le temps que je suis restée au

ailleurs, à tous les collègues doctorants et professeurs pour les bons moments de jeux,

x Acknowledgements

Maxime et Yohann pour les moments de partage sur le travail, mais aussi sur la vie en

général.

Je voudrais remercier le côté français du jury de thèse, Professor Eduardo Cursi et

remercie M. Xavier Brunotte de l'entreprise Vesta System pour avoir fourni une licence du

logiciel CADES pour l'exécution de ce travail. Je remercie aussi Frédéric Wurtz pour tout

le support administratif donné à moi et Tiago lors de notre arrivée et aussi pendant notre

séjour à Grenoble.

Merci à Rodrigo, Dyenny, William, Douglas, Angelica, Diego, Lucas, Guilherme, Paula,

Vinicius, Juliana, Marcelo, Thiago, Poliana, Vincent F., Jonathan et Savana, avec qui j'ai

À mes grands amis Pauline, Laura, Julie, Yohann, Lucie, merci pour tous les enseigne-

ments de français, sur la France, sur les montagnes, sur la vie! Vous êtes des personnes

incroyables et je suis chanceuse de vous avoir trouvé! Mon séjour en France était spéciale

à cause de vous!

Finally, for the nancial support, I would like to thank CAPES and the Region Rhône

Alpes.

List of Acronyms

DC Distribution Center

FA Factor Analysis

FR Fuzzy Reasoning

GI Goods Inwards

GP Global Performance

IT Information Technology

KF Kalman Filter

NS Normal Scale

xii List of Acronyms

OS Optimized Scale

PC Principal Component

Chapter 1

Introduction

mistakes which it is useful to make, because they lead little

by little to the truth.

Jules Verne

Contents

1.2 Research Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.2.1 Research Gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.2.2 Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.3 Dissertation Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.3.1 General Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.3.2 Specic Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.4 Methodology and Development . . . . . . . . . . . . . . . . . . . . . 8

1.5 Research Delimitations . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.6 Thesis Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

Abstract

This chapter presents the context of the study and the research gaps which are basis

for the work's objectives. Besides, the dissertation proposal is detailed, presenting the

research methodology and the steps carried out to achieve these objectives. Finally, the

research delimitations are discussed and the thesis structure is reported.

2 1. Introduction

The literature about performance measurement is vast. Performance measurement or or-

ganizational performance has become an important issue in companies due to the pressure

to give results (Kennerley and Neely, 2002). The performance indicators, which form the

performance measurement system, provide a tool to compare the current results with the

present objectives and thus to eventually launch the necessary actions to carry out in order

to reach these objectives (Berrah et al., 2000). Summarizing the literature of the last 30

years, it is possible to identify four main phases of the performance measurement area

(Neely, 2005).

First, in the 1980s it was the problem identication phase, where the dominant theme

Neely (2002) state that in this stage, there was a growing realization that, given the

increased complexity of organizations and the markets in which they compete, it was no

longer appropriate to use nancial measures as the sole criteria for assessing success. The

nancial measures are concerned with cost elements and quantify performance solely in

nancial terms, but many enhancements are dicult to quantify monetarily, such as lead-

time reduction, quality improvements and customer service (Tangen, 2004). So, there has

focus only on nancial results (Coskun and Bayyurt, 2008). The main reason is, according

to Fernandes (2006), that organizations compete not just on nancial eciency, but also

on social legitimacy. A company does not want just to maximize nancial revenues, but

By the early 1990s, the second phase potential solutions has proposed measurement

frameworks such as the balanced scorecard (Neely, 2005). Following these developments,

researches have started to suggest other performance measures, since nancial indicators

could not meet expectations of all stakeholders and a good organizational performance

should balance all organization dimensions which are related (Fernandes, 2006). Then, the

methods of application (third phase), involved the search for ways in which the proposed

Beginning of the 2000s was marked by the empirical investigation phase, in which peo-

ple have begun to look for more robust empirical and theoretical analysis of performance

measurement frameworks and methodologies. The objective was to develop dynamic rather

than static measurement systems and to ensure an appropriate focus on enterprise perfor-

mance management, rather than simply performance measurement (Neely, 2005). The

coordination. Alignment deals with the maintenance of consistency between the strategic

goals and metrics as plans are implemented and restated as they move from the strategic

Nowadays, we are in the information era. Internet has changed the way people and

companies relate to each other. This situation also has an impact in performance man-

agement methods. Lam et al. (2011) argue that information systems, such as warehouse

management system (WMS), are recognized as useful means to manage resources in the

warehouse. The information technology enables, for example, the product tracking from

1.1. Context of the study 3

of Things (IoT) is often considered to be part of the Internet of the future, consisting in

will have sensing, actuating, and often data-processing capabilities (Ng et al., 2013).

One of the changes coming from communication development is the conversion of local

ucts and services to satisfy customers while trying to reduce costs. It has led companies to

decentralize their production systems all over the world. So, supplying the correct product,

in the right time and in the right quantity has become a challenge, requiring a very good

management of all company areas. The logistics plays an important role by aggregating

value to the products and it has become a critical factor to obtain competitive advantages.

modeling and analysis become increasingly more important and dicult in the manage-

ment of such complex manufacturing logistics networks (Wu and Dong, 2007).

One of the important aspects under the responsibility of the logistics sector is the

warehouse, where the main logistics operations take place: transportation, warehousing

and stocking. Not only their number is increasing substantially but also their functionality

is changing. Whereas in the past many European Distribution Centers (EDCs) primarily

served as a warehouse with a distribution function, some of the current EDCs have Eu-

Koster and Waremius, 2005). The connection of these activities in one place makes the

performance measurement in the warehouse a key factor for the overall performance of the

logistics operations.

The growing warehouse operation complexity and the easy information access have led

dicult. The reason for that is the misunderstandings that managers could have when

assessing global warehouse performance, since dierent indicator characteristics make dif-

cult the evaluation of their structural relationships. Also, today managers are confronted

with greater uncertainty and unpredictability, complicating the decision making; wrong

Regarding the quantity of indicators used to manage performance, the managers have to

choose among a lot of indicators (having a complete set of informations to make decisions)

or few indicators (e.g the KPIs, Key Performance Indicators). In the rst case it is hard

to evaluate the global performance with so many data but, if the manager chooses few

indicators, the global evaluation is simplied and some important information can be lost.

In both cases, there will be indicators with dierent objectives (e.g. the level of a cost

indicator shall be minimized, while a quality indicator level shall be maximized). This fact

may increase the diculty of the analysis executed by the manager while evaluating the

warehouse global performance, even if he chooses a lot or few indicators. Cai et al. (2009)

conrm this conclusion arming that it is dicult to gure out the intricate relationships

among dierent KPIs and the order of priority for accomplishment of individual KPIs.

Nevertheless, even if managers would like to evaluate just few indicators, the more the

process is complex, the more the indicators needed are numerous and dierent (Melnyk

et al., 2004). Thus, the aggregation of indicators can considerably simplify the analysis of

4 1. Introduction

2006).

eective way on the global warehouse performance, considering the existing indicators of

the warehouse activities and knowing that there are limits in the decision-maker's ability

to process large sets of performance expressions (Clivillé et al., 2007). In this context, the

which aggregates indicators, giving a summarized feedback about the overall performance

tion, to the aggregation of operational indicators of the warehouse, since this area has the

Interestingly, the term performance aggregation has dierent meanings in the litera-

ture. For example, Böhm et al. (2007) state that performance information used at higher

decision levels is more aggregated than the one employed at lower levels due to various

reasons (data availability and error minimization, etc.). In this dissertation, we consider

order to achieve a measure, representing all the performance indicators of the system.

This denition is conrmed by Clivillé et al. (2007), who state that the aggregation of

The next section presents the literature supporting the research gaps which are fullled

by this dissertation.

This section is divided in two subsections. First, we present the research gaps reported

by previous works, explaining for which problems we propose solutions. Secondly, the

The literature on warehouse performance assessment has been largely ignored (Dotoli et al.,

2009; Johnson and McGinnis, 2011). While there are widely accepted benchmarks for indi-

vidual warehouse functions such as order picking, little is known about the overall eciency

of warehouses (Johnson and McGinnis, 2011). Gu et al. (2010) present a review about

design and performance evaluation of warehouses. The authors address important future

directions for the warehouse research community, stating that the total warehouse perfor-

we found very few papers analyzing warehouse performance relationships and proposing

frameworks to evaluate the global performance. The two main approaches used in the

First, Sohn et al. (2007) evaluate relationships among various inuential factors to

develop an Air Force Warehouse Logistics Index (WLI). This index evaluates the logistics

support capability of ROKAF (Republic of Korea Air Force) warehouses. The authors

1.2. Research Problem 5

The group of works in which Sohn et al. (2007) is included presents as the main charac-

teristic the acquisition of data from questionnaires in order to perform mathematical tools.

After interviewing people related to the subject, the papers can use several statistical tools

to conrm, or not, the proposed relationships. In most of the cases, the questionnaires do

not contain indicators' information and in the cases where there are indicators, they are

evaluated qualitatively.

The second approach evaluates the global warehouse performance without subjective

judgments. The papers use basically DEA (Data Envelopment Analysis) tool. For example,

Johnson et al. (2010) investigate the factors that impact warehouse performance (using

correlation method) and evaluate warehouses with regard to technical eciency (i.e. inputs

and outputs).

The DEA tool is usually used for benchmarking, and the database to perform it is

perfect orders (related to more than one activity) are not included in the model.

We observe that the literature on warehouse subject does not provide an aggregated

fore, we also verify the literature concerning the aggregation of performance measurement

Several authors discuss the aggregation of performance indicators and their relation-

ships.

whether the upstream objectives are being reached or not. However, no further informa-

tion about the causes is provided by these KPIs (Key Performance Indicators). For these

authors, the fact of discovering relationships between KPIs is potentially much more prof-

itable for an organization if it is possible to discover the latent relationships that occur

between objectives of the PMS. Then, cause-eect relationships between objectives could

be explained and managers would have additional decision-making information. For Mel-

nyk et al. (2004), while there are numerous examples of the use of various metrics, there

are relatively few studies in operations management that have focused on the eects of

metrics within either the operations management system or the supply chain.

Lauras et al. (2010) arm that each KPI should be examined separately and then in

related groups of indicators. Analysts such as the task leader or senior manager must

simultaneously consider all these factors. Regarding the number of indicators analyzed

simultaneously, Lohman et al. (2004) state that it is impossible for a manager to make

decisions on the basis of 100 unstructured metrics. Furthermore, Melnyk et al. (2004)

present the complexity of an individual's metrics set as a load imposed upon a person's

an aggregate indicator, which synthesizes the performance of the set of indicators. For

Vascetta et al. (2008), the aggregated indicator is an informative tool, able to provide

general background in a format that is easy to create and to update. In addition, it should

6 1. Introduction

have an attractive and understandable format to be considered helpful for people of all

Even if several authors have discussed the need of an aggregate measure, few works have

tried to accomplish it. Thus, the main research gaps which this dissertation proposes to

fulll are: Using a set of ratio measures can lead to confusion; if some measures are good and

some are poor, is the warehouse performing well? (Johnson et al., 2010). The challenge is

to design a structure to the metrics (i.e., grouping them together) and extracting an overall

sense of performance from them (i.e., being able to address the question of Overall, how

well are we doing?) (Melnyk et al., 2004). In the same way, Lohman et al. (2004) arm

that a conceptual question is still not answered: What are the eects of combining several

Even if some questions are asked more than 10 years ago, they are still valid since there

are a lot of developments to be made on this subject. One conrmation is the statement

of Clivillé et al. (2007), pointing out that as soon as managers use more than one KPI,

After the works of Melnyk et al. (2004) and Lohman et al. (2004), some papers have

studied ways to aggregate performance. These researches usually use a mathematical tool

based on manager's opinions or subjective judgments (e.g. Fuzzy, AHP - Analytic Hi-

erarchy Process) to achieve this objective (see, for example, Luo et al. (2010)). Also,

naires (see. Fugate et al. (2010)). Unlike these earlier works, we propose, in this disserta-

considering experience or subjective judgments inside the mathematical tools. For that,

analytical models and statistical tools are used to relate and aggregate indicators, including

The work of Rodriguez et al. (2009) is the closest we found to our proposition. Ro-

driguez et al. (2009) develop a methodology to dene aggregated indicators without judg-

ments, using the time series of indicators to measure their correlations and combine them

in factors. The main goal of the work is to relate the aggregated performance indica-

tors upstream towards the strategic objectives of the company, to analyze the objective

achievement.

This dissertation diers from Rodriguez et al. (2009) in the following points: our pur-

pose with the performance aggregation is to provide insights about warehouse performance

management in the operational level instead of strategical level; the application area of

our work is warehouses instead of enterprise administration; the statistical tool used by

Rodriguez et al. (2009) is just one part of the analysis performed in our work, since in this

dissertation, relationships are also determined analytically; we also develop a scale for the

The proposed work is relevant in the theoretical and practical points of view: this

subject has received less attention and this dissertation brings new insights about this

theme; companies can realize their global performance with the implementation of the

The complexity and diculty to get this solution to aggregate warehouse performance

1.2. Research Problem 7

1.2.2 Complexity

The complexity of this theme is addressed in dierent ways by the literature.

Caplice and She (1994) report some trade-os involving indicator's choice. One of

them is usefulness versus integration of indicators. This trade-o indicates that as a metric

becomes more aggregated it loses its direct usefulness. Moreover, if an indicator captures all

of the details of a process it tends to become more complex and thus harder to understand.

Franceschini et al. (2006) state that the eectiveness of an aggregated indicator strongly

depends on the aggregation rules, because sometimes its result can be questionable or even

misleading. Two years later, Franceschini et al. (2008) conrm that the aggregation of

several indicators into an aggregated indicator is not always easily achievable, especially

Vascetta et al. (2008) assert that the aggregation using mathematical equations nec-

essarily requires many assumptions and simplications which could lead to incorrect or

Beyond the strong criticism of indicator usefulness and the possible reluctance of man-

agers to utilize aggregated indicators, the main challenge is to provide trustful relationships

among indicators. We believe that once this last problem is solved, the others will be con-

siderably minimized. Thus, the proposition of this thesis is to relate indicators considering

just indicator equations and the time series of their results, without human judgments.

Two dierent quantitative methods (analytical model and statistical tool) are performed,

It is hard to model indicator relationships since several factors inuence their results.

De Koster and Balk (2008) exemplify this situation arming that common measures used

in warehouses (e.g. order lines picked per person per hour, picking or shipment error rates,

order throughput times) are not mutually independent and, additionally, each of them can

depend on multiple inputs. The result is that the indicators do not only inuence one

another (e.g. order lines picked per person per hour and order throughput time), but they

Another potential problem is how to provide a general solution with many dierent

kinds of warehouses. In this way, the rst issue is to dene the set of indicators to measure

warehouse performance. Clivillé et al. (2007) conrm that one major problem in the design

provide informations about the current warehouse situation and how far it is possible to

go. The complexity remains especially in the determination of the scale boundaries since

The next sections present the dissertation objectives and the development required to

8 1. Introduction

1.3.1 General Objective

The main goal of this dissertation is to develop a methodology for an integrated warehouse

To reach this objective, it is necessary to balance dierent indicators using mathematical

tools in order to consider the particularities of each of them. From the general objective

surement;

mance.

Each one of these specic objectives represents a contribution of this work. The next

The general research methodology applied in this dissertation is a quantitative model based

research. According to Bertrand and Fransoo (2002), this methodology is based on the

assumption that we can build models which explain (part of ) the behavior of real-life

operational processes or that can capture (part of ) the decision-making problems that are

Regarding the specic steps of this work, it is possible to dene two other sub method-

ologies. The rst one consists of a normative empirical quantitative research, dened as a

research in which policies, strategies and actions are developed (Bertrand and Fransoo,

2002). This methodology encompasses from step one of Figure 1.1 (Searches on Databases.

passing the rest of the work phases (Figure 1.1) corresponds to the descriptive empirical

research, which is primarily interested in creating a model that adequately describes the

causal relationships that may exist in reality, which leads to understanding the processes

1.4. Methodology and Development 9

of the work in three main columns: bibliographic research, development and outcomes.

The bibliographic research steps performed in the left column of Figure 1.1 are related

specically to the knowledge taken from the literature. This knowledge is used as a basis

for the development area (middle column). Finally, the outcomes are the results of the

developments carried out in this dissertation, also called the main contributions of the

work.

Definition and

Searches on Databases.

Determine the set of performance classification of

Keyword: warehouse

indicators and their definitions warehouse performance

performance

indicators

Definition of performance

indicators equations Normative

empirical

Searches on Databases.

Analysis of the literature about

quantitative

Keyword: Aggregate research

performance aggregation

performance/ indicator

Determination of a

Methodology to

Searches on Databases. Evaluation of statistical tools procedure to measure

determine an integrated

Keyword: statistical tools which allow to aggregate the integrated

warehouse performance

to reduce dimensionality indicators performance in

measurement

warehouses

Searches on Databases. Analysis of the literature about

Keyword: scale scale generation

Standard warehouse

definition

performance indicators and for indicator and data

the associated data equations (1)

the methodology (2) Descriptive

empirical

research

Correlation measurement among

Jacobian matrix assessment

indicators

Method to determine

Indicator’s aggregation using Develop a theoretical model

indicator relationships

statistical tools describing indicator relations

analytically

proposition model

Dissertation Conclusion

integrated model using (1) and (2) scale determination

Figure 1.1 starts with a deep literature review carried out in order to verify the set of

the literature does not provide a clear classication of these warehouse indicators regarding

their denitions. Thus, the rst outcome of this dissertation is the classication and

denition of the warehouse performance indicators. From this result, indicator denitions

on dierent themes are carried out to develop the methodology to determine an integrative

10 1. Introduction

subject and, also, discussing adequate statistical tools which should be used to aggregate

indicators and the way it should be made. To simplify the interpretation of the integrated

warehouse performance, it is also necessary to develop a reference scale to allow the evalua-

tion of performance results. These three themes (performance aggregation, statistical tools

and scale generation), together, structure the proposed methodology generating knowledge

in this area.

which contains the main processes/activities as usually found in real warehouses. The

performance indicators used for warehouse management are dened based on the literature

review ndings and an analytical model of indicator and data equations is generated (third

outcome).

To apply the mathematical tools and to nd indicator relationships a historical time

series of indicators is necessary. For that, data is generated representing the warehouse

dynamics with indicator results changing monthly. From these data, two dierent analysis

are performed to propose an integrated performance model (fth outcome). The rst anal-

ysis utilizes the analytical model and the data generated to verify indicator relationships

from the Jacobian matrix result. As a result of this development, we have the fourth out-

come, a method to determine analytically the indicator relationships. The second analysis

are analyzed carefully to determine the indicators which will make part of the integrated

performance model.

Finally, a scale is developed for the proposed integrated model. This scale is the result

of an optimization model which is based on the analytical equations of indicators and data

as well as the data generated to implement the methodology. As this kind of analysis

is quite new to design scales, the last outcome is the optimization model to dene the

integrated indicator scale. The conclusion of this dissertation with all developments return

The delimitations of this research and its results are divided in: methodology delimitations,

individual warehouse. It consists of the evaluation of one warehouse over a time period,

measuring its own performance periodically. Thus, this dissertation does not encompass

Secondly, the indicator set used in the standard warehouse (theoretical case) are taken

from the literature. In a real case, warehouses determine indicators from company goals.

As there are several developed frameworks to help managers with the indicators' choice

(e.g. Franceschini et al. (2008)), this dissertation does not address this subject. Thus, to

apply the methodology, it is considered that the selected indicators are the ones dened

by the company.

the results depend on warehouse characteristics and indicators. Even if there is no limi-

1.6. Thesis Structure 11

tation to use the methodology for indicators of higher levels or warehouses with dierent

characteristics, the method has not been tested/ veried on dierent applications in this

work.

The theoretical case study provides results that, initially, can not be generalized. The

numerical results obtained in the integrated performance model are limited by the consid-

erations made in data generation. However, regarding the analytical model, it is possible to

adapt it to similar warehouse situations, once the characteristics and operations presented

Regarding the metric set denition and indicator relationships, the delimitations are

as follows:

• The non-linear relationships among indicators are not measured in this work;

• Indicators for human resources performance measurement are not included in the

metric set. Only the indicators that relate persons to operations (e.g. productivity

• Indicators related to sustainable practices and reverse logistics activities are also not

Lastly, the numerical result of the developed scale can not be used in other warehouses

since to create it, it is necessary to dene low and high limits for data and indicators

according to the warehouse conditions. In this dissertation, some limits are dened based

on the restrictions proposed by the standard warehouse, as the maximum and minimum

number of products processed by the warehouse per month, whereas other limits are deter-

mined from indicator times series. However, the methodology to create the scale remains

a contribution of this work since its utilization is possible under the analytical model and

limits adaptation.

From this rst chapter which has presented the work proposition with its complexity and

tured method is used to classify papers and to obtain the main characteristics of the

literature concerning this subject. Furthermore, the indicators used for warehouse perfor-

mance assessment are acquired from papers and classied according to their dimensions of

measure.

and their relations. The main focus is to show papers using mathematical tools to assess

the global performance. These mathematical tools are classied and detailed, providing a

Chapter 5 describes the standard warehouse for which the performance measurement

is assessed. The warehouse activities, layout and the unit of measure of indicators are

12 1. Introduction

we generate a database for the theoretical warehouse, which will be used for illustration

purposes. After the database generation, the relationships among indicators are calculated

using the Jacobian matrix, correlation matrix and Principal Component Analysis.

Chapter 7 analyzes the results of the mathematical tools application and proposes an

integrated performance model. Also, a scale to evaluate the results of the integrated perfor-

mance model is developed and tested for two dierent warehouse performance situations.

Chapter 8 presents the conclusions from the work results, highlighting the main con-

Chapter 2

Literature Review on Warehouse Performance

possible de faire et tu réaliseras l'impossible sans t'en

apercevoir.

François d'Assise

Contents

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.2 Research methodology and delimitations . . . . . . . . . . . . . . 14

2.3 Results of Content Analysis . . . . . . . . . . . . . . . . . . . . . . . 17

2.3.1 Based on geographical and journal representation . . . . . . . . . . . . . 17

2.3.2 Based on the work methodology . . . . . . . . . . . . . . . . . . . . . . . 18

2.3.3 Application area of works . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.3.4 Warehouse activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.3.5 Warehouse Management tools . . . . . . . . . . . . . . . . . . . . . . . . 21

2.4 Direct Warehouse Performance Indicators . . . . . . . . . . . . . 23

2.4.1 Time related performance indicators . . . . . . . . . . . . . . . . . . . . . 24

2.4.2 Quality related performance indicators . . . . . . . . . . . . . . . . . . . 24

2.4.3 Cost related performance indicators . . . . . . . . . . . . . . . . . . . . . 25

2.4.4 Productivity related performance indicators . . . . . . . . . . . . . . . . 26

2.5 Indirect Warehouse Performance Indicators . . . . . . . . . . . 26

2.6 Classification of the Warehouse Performance Indicators . . 30

2.6.1 Specic and Transversal Indicators . . . . . . . . . . . . . . . . . . . . . 33

2.6.2 Resource Related Indicators . . . . . . . . . . . . . . . . . . . . . . . . . 34

2.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

Abstract

This chapter carries out a deep literature review on warehouse performance. We per-

form a descriptive analysis of selected articles using content analysis method. The

performance indicators acquired from these papers are divided initially as indirect or

direct indicators. The indirect indicators are rather related to concepts and there is

not a unique and simple equation to express them. The direct indicators are measured

by equations like ratios and are also classied according to the dimensions of time,

quality, cost or productivity. In order to clarify the direct indicators boundaries, we

provide a framework positioning the measures according to the activity and dimension

classication. Some conclusions made from this structured literature review are also

presented.

14 2. Literature Review on Warehouse Performance

2.1 Introduction

As the main objective of this dissertation is to study warehouse performance in an in-

tegrated way, a deep literature review is performed in this chapter to identify the main

this review synthesizes past works to recognize which kind of measures are mostly used

the literature, some classications are performed to organize them according to what they

measure (e.g. the performance of a specic activity) and how they do it (the mathematical

the warehouse performance, usually done in short and regular time intervals (like months).

These periodic results are used by managers to verify the evolution of the performance

along the time and to take actions to enhance better results. We refer to the performance

analysis as the measurement and comparison of actual levels of achievement with specic

objectives, measuring the eciency and the outcome of corporation (Lu and Yang, 2010).

In the following discussion, the terms metric, performance measure and performance

indicator are used as synonyms, as commonly done in the literature (Franceschini et al.,

2006).

The reviews found treating warehouse subjects address technical issues as storage ca-

pacity and assignment policies (Cormier and Gunn, 1992; Gu et al., 2007), order picking

problems (Cormier and Gunn, 1992; De Koster et al., 2007; Gu et al., 2007, 2010), routing

problems (De Koster et al., 2007; Gu et al., 2007), and layout design (Cormier and Gunn,

1992; De Koster et al., 2007; Gu et al., 2010). Only the work of Gu et al. (2010) addresses

The next sections present the methodology used for selecting and analyzing papers with

The process of collecting and selecting the papers is described in Figure 2.1. In the Initial

Search phase, we dened a list of relevant keywords used for the database search, as

demonstrated by the three parts of Table 2.1. The rst subtable in the left side of Table

2.1 demonstrates the databases researched and the subtable in the right side shows the main

keywords utilized. The third subtable of Table 2.1 (located below the rst two subtables)

presents all 24 possibilities of Keywords Combinations tested in all databases. The initial

search did not limit publication year and document type; the only limitation was the results

published in available English-language. This initial search resulted in 1500 articles, where

1090 were from journals and 410 from conferences, magazines and reports. We focus on

Analyzing the article's publication year, we found that the rst publication about

warehouse performance appears in 1970's with the work of Lynagh (1971). But the number

of relevant papers available in databases up to 1990 is really rare. We can cite just the works

of Khan (1984) and Svoronos and Zipkin (1988) as examples. To be sure that the literature

review contains the majority of articles during a range of years, this study was restricted

2.2. Research methodology and delimitations 15

Databases Keywords

Scopus (scopus.com) Warehouse/ Distribution Center

Emerald (emeraldinsight.com) Facility Logistics/ Logistics Platforms

EBSCO (ebscohost.com) Performance/ Eciency

Wiley (onlinelibrary.wiley.com) Evaluation/ Measurement/ Assessment

Science Direct (sciencedirect.com) Logistics/ Logistics audit

Web of Science (webofknowledge.com) Operation Management

Compendex (engineeringvillage2.com) Metrics/ index/ KPI

Keyword combinations

Performance Measur∗ / Assessment & warehouse / distribution center / logistics platform

Performance Measur∗ / Assessment & warehous* / DC & logist∗

Performance Evaluation & distribution center / logistics platform

warehouse/ distribution center / logistics platforms & performance

warehouse operations management

warehouse / distribution center & logistics index

warehouse eciency & measur∗

performance & metric & warehouse

warehouse overall performance

warehouse management & logistics

warehouse & logistics KPI

logistics performance

on publications from 1991 up to 2012. This range of years oers sucient support to make

Following the steps presented in Figure 2.1, in the third phase, the journals articles

are ltered by considering that their titles contain the keywords: (i) warehouse or similar

(Distribution Center, Facility Logistics, Logistics Platforms, Cross Docking); (ii) the words

performance or management or evaluation and the warehouse area / activity; (iii)

logistics management and logistics performance measurement. During this selection, review

papers in the warehouse area are also considered. From this stage, the database is narrowed

Finally, the abstract of each article is analyzed. In this phase, the papers are ltered

content, the full text was also veried. Note that the nal database (43 articles) does not

16 2. Literature Review on Warehouse Performance

mentation;

• Warehouse design;

• Warehouse location;

• Distribution optimization.

The justication of not including the subjects cited above is that they treat strategical

and tactical decision making (e.g. warehouse location, design) and not the operational

performance management which is the main focus of our literature review (e.g. unloading

Only the works using decision making for operational warehouse management are taken

into account. As the decision support tools are considered as means to manage the perfor-

mance, the articles presenting decision support systems (DSS) to help warehouse manager's

decisions (Lam et al., 2011; Lao et al., 2011, 2012) and the articles treating the system

inuence on enterprise performance (Autry et al., 2005; Karagiannaki et al., 2011) were

The nal database is used to make two dierent analysis as shown in Figure 2.2. First,

we provide a descriptive analysis of all 43 papers in Section 2.3. That is, a quantitative

evaluation of the general characteristics of the articles. The second analysis, presented in

Section 2.4 and 2.5, focuses on the performance indicators used in warehouses. In the nal

database, only 35 articles present performance indicators. Among these 35 papers, 32 arti-

cles discuss the performance indicators which can be expressed by some simple equations,

being measured directly. We qualify them as direct indicators. We address this kind of

papers in Section 2.4. There are 16 articles among 35 that assess performance indicators in

an indirect way. It means that these indicators represent more complex concepts which

are dicult to measure by simple expressions like ratios. Therefore, more sophisticated

statistical tools (e.g. regression analysis) are used to assess them. These performance in-

dicators are named as indirect indicators and an analysis of them is provided in Section

2.5.

The papers of the nal database are explored based on content analysis research

atically evaluate the literature in terms of various categories, transforming original texts

Content analysis can be carried out in two steps: denition of variables analyzed and

the unitization of them. The denition of the variables depends on research objectives. In

this dissertation, the variables extracted from papers are: work methodology, mathematical

tools utilized, warehouse activities and indicators used to assess performance. The second

systematic distinguishing of segments of text that are of interest to an analysis. That is,

2.3. Results of Content Analysis 17

in the nal paper database we look for the variables and when they are not explicit in the

text some predened rules are used to classify the information acquired from the text. In

order to maintain consistency in this procedure and to avoid biases, this step is conducted

by the author of this thesis (this procedure is usually adopted when performing content

analysis according to Krippendor (2004)). This principal reader has lled the variables as

presented in each study on a spreadsheet. This master listing of ndings is then analyzed

The results of the spreadsheet analysis are given in the next sections.

This section shows the content analysis by using tables which present some quantitative

outcomes resulted from paper's classication. They present patterns identied from the

data, allowing to categorize the warehouse performance literature. More specically, Sec-

tion 2.3.1 shows the number of publications per continent and per journal, Section 2.3.2

introduces paper methodologies, Section 2.3.3 shows their classication by application ar-

eas, Section 2.3.4 presents the warehouse activities most studied in the works and Section

2.3.5 summarizes the tools developed for helping managers on warehouse management.

Figure 2.3 shows one of the results from article analysis, the number of publications over

years per continent. We note that the sum of the number of publications per continent/year

could be more than the total curve value because some papers are co-authored by people

from dierent continents and are counted more than once. From Figure 2.3 several infer-

ences could be made. First, it is apparent that research on warehouse performance has

increased in the last years, demonstrating the subject relevancy. Second, the represen-

tation of European papers has also increased substantially in the last years. The main

European publishing countries are The Netherlands, Greece and Italy with four, three and

three publications each, respectively. America, on the other hand, maintains almost the

same number of publications over years with United States being the country with most

publications (16 papers) of all continents. Third, the number of papers realized in interna-

tional cooperation sums to 10 publications, almost one fourth of our database. Europe is

the continent with the highest international co-authoring (7 papers), followed by America

(6 papers).

18 2. Literature Review on Warehouse Performance

Journal NP a

%

European Journal of Operational Research 5 11.6

Journal of Business Logistics 3 7.0

Journal of Manufacturing Technology Management 3 7.0

International Journal of Production Research 2 4.7

International Journal of Production Economics 2 4.7

TOTAL 15 35

a

Number of publications

addressed, Table 2.2 demonstrates the journals that most publish in the area. The re-

sults show that publications are very widespread since the journals with one publication

represent more than 60% of the selected articles. So, we can conclude that this area is

very interdisciplinary. The "European Journal of Operational Research" has the highest

concentration with ve articles. It is interesting to highlight that four among these ve

publications are literature reviews showing the general interest on this subject area.

Other data points acquired by the descriptive analysis capture the articles' methodology.

The articles are classied based on ve research methodologies (see Seuring and Muller

(2008)): mathematical, conceptual, case study, survey, and review papers. A paper is

standard deviation, etc.) as well as more sophisticated tools (e.g. linear regression, ana-

lytical model, simulation) are used. To be classied as conceptual, the work needs to be

mented in practice. The case study is a work that develops a theory and veries the results

a research paper carrying out a questionnaire to make conclusions about a subject. Each

paper could be classied in more than one methodology, depending on its characteristics.

The exception is the review papers, which were separated because of their relevance. The

2.3. Results of Content Analysis 19

Mathematical

Case Study

Conceptual

Review

Survey

NPa % Articles

X X 17 39.5 Kiefer and Novack (1999); Ellinger et al. (2003); Autry et al.

(2005); De Koster and Waremius (2005); Voss et al. (2005);

Sohn et al. (2007); De Koster and Balk (2008); Park (2008);

O'Neill et al. (2008); Menachof et al. (2009); Forslund and Jon-

sson (2010); Lu and Yang (2010); De Marco and Giulio (2011);

Johnson and McGinnis (2011); Markovits-Somogyi et al. (2011);

Banaszewska et al. (2012); Yang and Chen (2012)

X X 13 30.2 Wu and Hou (2009); Manikas and Terry (2010); Matopoulos

and Bourlakis (2010); Wang et al. (2010); Johnson et al. (2010);

Cagliano et al. (2011); Lam et al. (2011); Goomas et al. (2011);

Karagiannaki et al. (2011); Lao et al. (2011); Sellitto et al.

(2011); Lao et al. (2012); Ramaa et al. (2012)

X 5 11.6 Cormier and Gunn (1992); van den Berg and Zijm (1999); De

Koster et al. (2007); Gu et al. (2007, 2010)

X 3 7.0 Spencer (1993); Gunasekaran et al. (1999); Gallmann and

Belvedere (2011)

X 3 7.0 Mentzer and Konrad (1991); Rimiene (2008); Bisenieks and

Ozols (2010)

X X 2 4.7 Yang (2000); Saetta et al. (2012)

TOTAL 43 100.0

a

Number of publications

The quantitative works represent 74.4% of the total papers (i.e. survey/mathematical

their signicance, we detailed the quantitative works according to the type of method used

(see Table 2.4). The basic statistics are further detailed as ANOVA and F test; p value and

σ; and Others. We note that some papers use more than one mathematical tool. In such

papers, most of the time, the basic statistics are combined with other tools. For example,

factor analysis or regression analysis are combined with the basic statistics to describe

relations among warehouse activities (10 out of 32 papers). Another example is the use of

statistics to compare the simulation results. The next subsections present which kind of

industries and warehouse activities were most representative according to the database.

To verify the most relevant application areas, we classify the articles based on the position

of the application point in the supply chain. Table 2.5 shows three major classes as:

(1) manufacturing industries (with their respective Distribution Centers - DC). In this

category, the articles are further classied as one industry and as several industries if the

application is on a single or on several industries, respectively; (2) retailers, and (3) third

party logistics. We classify as Other the works which are not related to any industrial

activity, like Air Force (see Sohn et al. (2007)) and as Not Specied if application areas

are not mentioned. The main area appearing in papers is the Food industry, with a total

20 2. Literature Review on Warehouse Performance

Math Tool NPa %

(1) Basic Statistics 20 40 Area NPa %

(1.1) ANOVA and/or F test 8 16

(1) Manufacturer and its DC 19 44.2

(1.2) σ b , p value 7 14

(1.1) One industry 6

(1.3) Others 5 10

(1.2) Several industries 13

(2) Regression Analysis 6 12

(2) Retailers 9 20.9

(3) Factor Analysis 5 10

(3) Third Party Logistics 6 14.0

(4) DEAc 5 10

(4) Other 1 2.3

(5) Analytical Model 4 8

(5) Not Specied 8 18.6

(6) Simulation 4 8

Total 43 100

(7) Others 6 12 a

Number of publications

Total 50 100

a

Number of publications b standard deviation

c

Data Envelopment Analysis

presented in Table 2.5 show that 13 out of 19 articles related to a manufacturing domain

cover several industries. This is not very surprising when we cross check with Table 2.3.

We observe that there are a lot of survey papers (see Table 2.3) providing performance

comparison among enterprises. Such articles analyze dierent industry segments at the

same time.

We have also analyzed the kind of facility studied in the selected articles (warehouse

or distribution center (DC)), but it is dicult to provide reliable statistics on this subject.

Even though Manikas and Terry (2010) highlight that main dierences exist between these

two, dening aDC as a warehouse that emphasizes the rapid movement of goods , the same

authors also state that a distribution center could be similar to a warehouse in terms of

layout and operations management . In fact, in the related literature the terms DC and

warehouse are often used as synonyms (van den Berg and Zijm, 1999; Dotoli et al., 2009).

Therefore, in this work, we consider all indicators and management practices realized in

Warehouses could have dierent activities according to product specication, customer re-

quirements and service levels oered. For De Koster and Waremius (2005), the complexity

of the warehouse activities depend mainly on: (i) the number and variety of items to be

handled; (ii) the amount of daily workload to be done; and (iii) the number, the nature

and the variety of processes necessary to fulll the needs and demands of the customers

and suppliers.

Even though dierences may exist among the warehouse activities, they were dened

as: receiving, storage, order picking and shipping (van den Berg and Zijm, 1999). In

what follows we will use this generic classication. Some studies related to warehouse

performance also mention the delivery process (5 articles are identied). In some cases,

the delivery could be considered as a warehouse responsibility in the metrics sense. This

However, we did not include other warehouse activities such as replenishment (transfer

of products from the reserve storage to the picking area (Manikas and Terry, 2010)) and

sorting (if the picking is performed in batches, the products could be sorted before packing)

2.3. Results of Content Analysis 21

in this analysis because the database papers do not present performance indicators for these

activities.

As each of these ve activities can be divided into several sub-activities, we consider

• Receiving: operations that involve the assignment of trucks to docks, the scheduling

• Storing: material's movement from unloading area to its designated place in inventory

• Order Picking: process of obtaining a right amount of the right products for a set

of customer orders (De Koster et al., 2007). This is the main and the most labor-

• Shipping: execution of packing and truck's loading after picking, involving also the

• Delivery: the transit time for transportation from the warehouse to the customer.

Based on the above warehouse activities, the selected articles are analyzed and classi-

ed as in Table 2.6. This table helps identifying the major research areas by warehouse

activities.

A major observation we make out of Table 2.6 is that almost 40% of the articles

consider all major activities of the warehouse at the same time (rows 1 and 2 of Table

2.6). The articles mentioned in the second row (except Mentzer and Konrad (1991)) are

on the employee performances. According to van den Berg and Zijm (1999) and Mentzer

and Konrad (1991), the labor tasks impact all warehouse activities. Therefore, we choose

Another interesting insight is the fact that the majority of the articles include the

picking activity in their studies. This is quite relevant with industrial observations and

shows a certain maturity in the works undertaken. The order picking process is the most

costly among all warehouse activities, because it tends to be either very labor intensive

(manual picking) or very capital intensive (automatic picking). More than 60% of all

operating costs in a typical warehouse can be attributed to order picking (van den Berg

In the nal database, we nd some works which explore warehouse management systems

for decision aid and performance management. As these warehouse management tools are

The early works on warehouse management are rst focused on examining the processes

and identifying areas where an ecient management could improve the performance of

the warehouse. For example, Spencer (1993) presents a method based on value-added tax

(V-A-T) analysis and Theory of Constraints (TOC) to identify such critical process points;

Gunasekaran et al. (1999) study the problems in Goods Inwards (GI) area and provide

22 2. Literature Review on Warehouse Performance

Receiving

Shipping

Delivery

Storage

Picking

NP a

% Articles

X X X X 12 27.9 Cormier and Gunn (1992); van den Berg and Zijm (1999);

Gunasekaran et al. (1999); Kiefer and Novack (1999); Gu

et al. (2007); Rimiene (2008); Karagiannaki et al. (2011);

Cagliano et al. (2011); Gallmann and Belvedere (2011); Lao

et al. (2012); Yang and Chen (2012); Ramaa et al. (2012)

X X X X X 5 11.6 Mentzer and Konrad (1991); Ellinger et al. (2003); Wu and

Hou (2009); Lu and Yang (2010); Sellitto et al. (2011)

X X 5 11.6 Spencer (1993); Autry et al. (2005); De Koster and Balk

(2008); Johnson et al. (2010); Johnson and McGinnis (2011)

X X X 3 7.0 De Koster and Waremius (2005); O'Neill et al. (2008);

Saetta et al. (2012)

X 3 7.0 De Koster et al. (2007); Lam et al. (2011); Goomas et al.

(2011)

X X 2 4.7 Bisenieks and Ozols (2010); Gu et al. (2010)

X X X 2 4.7 Manikas and Terry (2010); Wang et al. (2010)

X X X 2 4.7 Menachof et al. (2009); De Marco and Giulio (2011)

X 2 4.7 Sohn et al. (2007); Park (2008)

X X X 1 2.3 Matopoulos and Bourlakis (2010)

X X 1 2.3 Voss et al. (2005)

X 1 2.3 Forslund and Jonsson (2010)

X 1 2.3 Markovits-Somogyi et al. (2011)

X X X 1 2.3 Banaszewska et al. (2012)

X X 1 2.3 Yang (2000)

X X 1 2.3 Lao et al. (2011)

Total 43 100.0

a

Number of publications

solutions to increase the performance of warehousing operations using Just in Time (JIT)

and Total Quality Management (TQM). These early techniques do not necessarily need

ticated IT tools in warehouses and DCs. Since 2000, more complicated algorithms and

ticles follow the same trend and propose utilization or development of decision support

mation systems, such as Warehouse Management System (WMS), are recognized as useful

means to manage resources in the warehouse (Lam et al., 2011). Wang et al. (2010) pro-

pose a Digital Warehouse Management System (DWMS) based on RFID (Radio Frequency

Identication) to help managers achieve better inventory control, as well as to improve the

operation eciency. Cagliano et al. (2011) model the warehouse processes using System

Dynamics and develop a dynamic decision support tool to assign employees to counting

tasks. Lam et al. (2011) develop a Decision Support System (DSS) to facilitate warehouse

order fulllment: when there is an incoming customer order, previous similar cases are

retrieved as a reference solution to the new incoming order. Lao et al. (2011) develop a

real-time inbound decision support system with three modules, which integrate the RFID

2.4. Direct Warehouse Performance Indicators 23

technology, Case-Based Reasoning (CBR), and Fuzzy Reasoning (FR) techniques to help

monitor food quality activities. Lao et al. (2012) propose a RFID based system to facilitate

the food safety control activities in receiving area, by generating a proper safety plan.

To evaluate the technology investment, Autry et al. (2005) design a method to deter-

outcomes for the warehouses or not. More recently, Yang and Chen (2012) and Ramaa

et al. (2012) study the impact of information systems on warehouse performance. These

studies conclude that the introduction of new technologies like RFID and WMS permits the

integration of decision support tools in warehouse management and improves the manager's

decisions.

In the next section, we present further analysis on the selected articles. But this time,

the analysis is focused more specically on the indicators used to assess the warehouse

performance.

The traditional logistics performance measures include hard and soft metrics. The rst

one treats quantitative measures such as order cycle time, ll rates and costs; the second

deals with qualitative measures like manager's perceptions of customer satisfaction and

loyalty (Chow et al., 1994; Fugate et al., 2010). The "hard" metrics are computable with

some simple mathematical expressions while the soft metrics require more sophisticated

tools of measurement( e.g. Regression analysis, fuzzy logic, Data Envelopment Analysis,

etc.). This work will refer to the "hard" metrics as direct indicators and the soft ones as

indirect indicators. The rst group will be presented in this section, and the second one

For the purpose of the analysis, all direct indicators are extracted from papers and clas-

These are: time (Mentzer and Konrad, 1991; Spencer, 1993; Neely et al., 1995; Frazelle,

2001; Chan and Qi, 2003; Gunasekaran and Kobu, 2007; Gallmann and Belvedere, 2011),

quality (Neely et al., 1995; Stainer, 1997; Frazelle, 2001; Gallmann and Belvedere, 2011),

cost (Neely et al., 1995; Mentzer and Konrad, 1991; Beamon, 1999; Chan and Qi, 2003;

Cai et al., 2009; Keebler and Plank, 2009), and productivity (Stainer, 1997; Frazelle, 2001;

Chan and Qi, 2003; Keebler and Plank, 2009; Gallmann and Belvedere, 2011). We note

that; some works prefer to use exibility instead of productivity as the fourth dimension

(Neely et al., 1995; Stainer, 1997; Beamon, 1999; Gunasekaran and Kobu, 2007), dening

nasekaran and Kobu (2007) state that exibility may be intangible and dicult to measure

in some cases. We present in Section 2.5 that exibility is preferably measured indirectly

rather than directly. Consequently, in this section productivity will be used as a dimension

The following procedure is used for the classication. Initially, all the direct indicators

found in the selected papers are listed. Once the list is completed, two types of aggregations

are made: (i) similar indicators are regrouped; (ii) very specic metrics are included in more

generic ones. One example of this second group is the work by Manikas and Terry (2010)

mentioning the indicator time of quality control in receiving. This can be considered as a

24 2. Literature Review on Warehouse Performance

portion of the receiving operation time; we include this indicator together with the class

of indicators called the receiving operation time. Finally, the indicators are organized

according to what they measure (time, quality, cost or productivity). We note that, for

the sake of uniformity throughout this work, the classications presented here are based

on our interpretation, instead of the original category proposed by the selected papers. For

example, Banaszewska et al. (2012) consider the number of consignment processed per

indicator. In this review we propose a sub-category, called the labor productivity and

Banaszewska et al. (2012) appears in this (see Table 2.10). Another example is the article of

Saetta et al. (2012), where the authors measure the customer satisfaction as the percentage

of orders on time and we classify the article under a broader indicator which is the on

time delivery (see Table 2.8). The classications resulting from this analysis are given

in Tables 2.7, 2.8, 2.9 and 2.10. We present a discussion on each class in the following

sections.

Table 2.7 shows the results for time related indicators. The most used metrics are order

lead time, receiving operation time and order picking time, respectively. Surprisingly, order

picking time is in the third position, even though Gu et al. (2007) state that past research

has focused strongly on order picking since this activity has large impact on the warehouse

performance. One reason could be that in the literature, the order picking time is more

specically treated in optimization works, which are not considered in this review.

Analyzing the time spent by a product in the warehouse through all activities, the

indicators found in Table 2.7 encompass almost all time components (receiving, putaway,

picking, shipping and delivery). The exceptions are the replenishment and inventory time:

there is no paper using an indicator like inventory coverage or replenishment time to mea-

sure it. Mentzer and Konrad (1991) presents indicators covering most of the activities in

author has measured the entire time spent by a product in the warehouse (since receiving

Regarding the warehouse activities covered by indicators, for the inbound processes

there are receiving and putaway times and for outbound processes picking, shipping and

delivery times. Interestingly, these ve indicators could be represented by just two: dock

to stock time (for inbound process) and order lead time (for outbound process). In the

case of order lead time, this indicator comprehends also administrative time beyond the

activities presented (picking, shipping and delivery) since its denition expresses, according

to Kiefer and Novack (1999), that order lead time starts to be measured at the time the

Dierent from the time dimension, the quality embraces measures linked with customer

The Table 2.8 illustrates the indicators used in the selected papers. We observe that

the emphases are on on-time delivery, customer satisfaction and order ll rate. The

2.4. Direct Warehouse Performance Indicators 25

Equipment downtime

Delivery Lead Time

Order picking time

Order lead time

Receiving time

Putaway time

Queuing time

Loading time

Authors

Mentzer and Konrad (1991) X X X X X X

Kiefer and Novack (1999) X

Yang (2000) X

Gu et al. (2007) X X X X

O'Neill et al. (2008) X

Rimiene (2008) X

Menachof et al. (2009) X

Manikas and Terry (2010) X

Matopoulos and Bourlakis (2010) X X

Wang et al. (2010) X

Cagliano et al. (2011) X X X

Lam et al. (2011) X

Gallmann and Belvedere (2011) X

Karagiannaki et al. (2011) X

Lao et al. (2012) X

Yang and Chen (2012) X X

Ramaa et al. (2012) X X

Total/each indicator 9 5 4 3 3 2 2 1 1

result corroborate with the statement of Forslund and Jonsson (2010), that perfectorder

results supplier delivery performance in a more comprehensive way, but seems not to be as

widely applied as on-time delivery .

The inventory, the warehouse physical area in which the products remain until they are

performance. Gallmann and Belvedere (2011) state that companies take into account

inventory management as a key to reach excellent service levels. Although inventory is not

an activity, its indicators (represented in Table 2.8 by Physical inventory accuracy) were

The results for cost dimension are presented in Table 2.9. It is interesting to note that

fewer works are recorded for cost indicators compared to the other dimensions. It could be

explained by the armation of Gunasekaran and Kobu (2007) that the operational level

on company's characteristics and choices. Despite the strategic importance in the supply

Table 2.9 also shows that the majority of the works mentioning cost metrics use in-

ventory cost indicator. From this data, it is apparent that what really interests managers

regarding the warehouse management costs is the inventory. The inventory is a cost gen-

erator by nature: according to Kassali and Idowu (2007), inventory is a business that

26 2. Literature Review on Warehouse Performance

b

a

Ord. ship. on time

Shipping accuracy

Delivery accuracy

Cust. satisfaction

Storage accuracy

Picking accuracy

On-time delivery

Perfect Orders

Stock-out rate

Order ll rate

Scrap rate

Authors

Mentzer and Konrad (1991) X

Gunasekaran et al. (1999) X X

Kiefer and Novack (1999) X X X X X

De Koster and Waremius (2005) X

Voss et al. (2005) X X X X X

De Koster and Balk (2008) X

Rimiene (2008) X X X

Menachof et al. (2009) X

Forslund and Jonsson (2010) X

Lu and Yang (2010) X X

Wang et al. (2010) X

De Marco and Giulio (2011) X

Lam et al. (2011) X

Gallmann and Belvedere (2011) X

Johnson and McGinnis (2011) X

Lao et al. (2011) X X X

Banaszewska et al. (2012) X X

Lao et al. (2012) X X X

Saetta et al. (2012) X X

Yang and Chen (2012) X X X X X X X

Ramaa et al. (2012) X X X X X

Total/each indicator 10 8 7 5 4 3 2 2 2 2 2 1 1

a

Customer satisfaction b

Orders shipped on time c

Physical inventory accuracy

involves costs and risk. The risks may come from probable product losses (e.g. quality

Another important dimension for the warehouse management is the productivity. Produc-

tivity can be dened as the level of asset utilization (Frazelle, 2001), or how well resources

are combined and used to accomplish specic, desirable results (Neely et al., 1995).

It can be seen from Table 2.10 that labor productivity and throughput are the most

employed metrics in warehouses. This result reinforces the fact that these are the main

In the past, the distribution centers (DC) primarily served as warehouses with distribu-

tion functions. Nowadays, the DCs have international headquarters, call-centers, service

centers or even manufacturing facilities as well (De Koster and Waremius, 2005). This

evolution is the outcome of a need to provide tailored services for the customers and to

2.5. Indirect Warehouse Performance Indicators 27

Cost as % of sales

Maintenance cost

Distribution cost

Inventory cost

Labor cost

Authors

Mentzer and Konrad (1991) X

Kiefer and Novack (1999) X

Yang (2000) X X

Ellinger et al. (2003) X

Rimiene (2008) X X

Johnson et al. (2010) X

Lu and Yang (2010) X X X

De Marco and Giulio (2011) X

Cagliano et al. (2011) X X

Gallmann and Belvedere (2011) X

Saetta et al. (2012) X

Ramaa et al. (2012) X X

Total/each indicator 7 3 2 2 2 2

gain competitive advantage. These new services require additional indicators to measure

the related performance. Oftentimes, the indicators are complex; either the equations are

not available or they are too dicult to calculate. The warehouse capability (Sohn et al.,

2007), the supervisory coaching behavior (Ellinger et al., 2003), the relation between front-

line employee performance and interdepartmental customer orientation (Voss et al., 2005),

etc. are some examples of these indicators. In this dissertation, we call such indicators,

the indirect indicators. Instead of simple and straightforward equations, some structured

mathematical tools are needed to calculate the value of these indicators. Normally, these

mathematical tools evaluate dierent kinds of information and extract correlations and/or

performances from databases. Some examples of such tools used in the literature are: SEM

Canonical Matrix.

The papers presenting indirect indicators are listed in Table 2.11. We give next some

characteristics in order to assess the capability of each warehouse taking part in the

study. The facility management is determined by the authors as: (i) maintenance

(iii) new construction of modern warehouses, and (iv) full equipment for protecting

facilities against re. As a result of the study, Sohn et al. (2007) conclude that facility

management.

• Flexibility: we can verify that the exibility measures are usually associated with

other performance components such as time, volume, delivery. For example, Lu and

28 2. Literature Review on Warehouse Performance

Receiving productivity

Outbound space util.b

Shipping productivity

Warehouse utilization

Transport utilization

Picking productivity

Labor productivity

Throughput

Turnover

Authors

Mentzer and Konrad (1991) X X X X

Gunasekaran et al. (1999) X

Kiefer and Novack (1999) X X X

De Koster and Waremius (2005) X

Voss et al. (2005) X

Gu et al. (2007) X

De Koster and Balk (2008) X X

O'Neill et al. (2008) X X X

Rimiene (2008) X X X X X

Johnson et al. (2010) X X X X

Manikas and Terry (2010) X X X

Matopoulos and Bourlakis (2010) X X X

Wang et al. (2010) X X

De Marco and Giulio (2011) X X X

Cagliano et al. (2011) X

Goomas et al. (2011) X

Johnson and McGinnis (2011) X X X

Karagiannaki et al. (2011) X

Markovits-Somogyi et al. (2011) X

Banaszewska et al. (2012) X X X

Yang and Chen (2012) X

Ramaa et al. (2012) X X X

Total/each indicator 11 11 8 4 4 3 3 3 2 1

a

Inventory space utilization b

Outbound space utilization

customer requests, delivery time exibility and volume exibility. Yang and Chen

(2012) consider exibility as urgent order handling and De Koster and Balk (2008)

consider exibility as the capacity to cope with the internal and external changes.

• Labor: the results in Table 2.11 demonstrate the importance of employee perfor-

mance in warehouses with numerous articles in the area. Ellinger et al. (2003) inte-

the supervisors). Voss et al. (2005) show that the front-line employee performance

their study, the authors consider the following variables to measure the employee per-

formance: proper data recording, ecient trailer loading, storing products in proper

damage, high productivity, high performance. Wu and Hou (2009) propose a model

for the analysis of employee performance trends. This model is intended to determine

the employees to reward or to train. Goomas et al. (2011) evaluate the order selec-

2.5. Indirect Warehouse Performance Indicators 29

a

VAL activity

Maintenance

d

Cust. Perc.

Inv. Man.c

War. Aut.

Flexibility

Labor

Authors

Kiefer and Novack (1999) X

Ellinger et al. (2003) X

Voss et al. (2005) X

Sohn et al. (2007) X X X

De Koster and Balk (2008) X X X X X

Park (2008) X

O'Neill et al. (2008) X

Wu and Hou (2009) X

Lu and Yang (2010) X X X X

Wang et al. (2010) X

Gallmann and Belvedere (2011) X

Goomas et al. (2011) X

Johnson and McGinnis (2011) X

Banaszewska et al. (2012) X X X

Yang and Chen (2012) X X

Total 7 4 4 4 3 3 1

a

Customer Perception VAL - Value Added Logistics Inventory Manage-

b c

the number of completed tasks, the number of tasks in queue and the team perfor-

mance against the engineered labor standards. Park (2008) study the relationship

between the store-level performance and the composition of the workforce. Workforce

(2010). Accordingly, Kiefer and Novack (1999) state that understand the inuence

of some measures in customer's reaction is far more important than any internal

measure alone.

De Koster and Balk (2008) measure customer perception by using DEA. The authors

verify the contribution of some activities (like cross-docking, cycle counting, return

VAL activities oered by the company and performed in warehouses. De Koster and

Balk (2008) divide VAL activities in low and high levels. The activities adding low

value to the product include labeling, putting manuals, kitting; whereas high VAL

30 2. Literature Review on Warehouse Performance

For Gu et al. (2007), the roles of VAL activities also include: buering the material

ow along the supply chain to accommodate variability caused by factors such as

of products from various suppliers for combined delivery to customers. The survey

of O'Neill et al. (2008) conrm that VAL activities have become common activities

in warehouses. However, on the average only 5 per cent of oor area is dedicated to

• Inventory Management: is an area where the automation support for activities has

are getting closer to each other. Wang et al. (2010) propose a digital warehouse

inventory control. Yang and Chen (2012) examine the impact of information systems

on DC's performance. Among the results, they found a positive correlation between

warehousing and inventory management and emergent order handling. In Sohn et al.

(2007), the issues related to the inventory management and the accuracy of logistics

• Warehouse Automation: De Koster and Balk (2008) measure the degree of ware-

WMS are low levels; RFID and barcoding or robots are high levels). Banaszewska

information systems.

The impact of the use of warehouse automation on its performance has also been

addressed. Yang and Chen (2012) conclude that high levels of information systems

utilization in the order selection activity should have positive inuences on delivery.

Throughout the classication process of direct indicators, we have observed that it is neither

easy to draw straight forward frontiers for them, nor are the measurements clearly dened.

For example, we could see two indicators with dierent names but measured the same way.

Conversely, some metrics have the same name but measured dierently. Moreover, while

in some papers the measurements are explicit, in some others only the indicator names are

given.

In order to provide well dened boundaries for the direct warehouse indicators, the

results presented previously in this chapter are analyzed using an activity-based frame-

work. The indicators that are classied in Section 2.4 according to quality, cost, time and

productivity dimensions, are now also classied in terms of warehouse activities described

in Section 2.3.4. The result of this new classication is illustrated by Table 2.12.

In order to classify the direct indicators with respect to the warehouse activities, we

• Transversal Indicators: are dened for a process rather than a unique activity. There-

2.6. Classication of the Warehouse Performance Indicators 31

• Resource related Indicators: Some indicators are related to the resources used in

the warehouses. We divide them into two distinct categories: Labor and equip-

ment/building.

32

Table 2.12: Direct indicators classied according to dimensions and activities boundaries.

Receiving Storing Inventory Picking Shipping Delivery

receiving putaway order pick- shipping delivery lead

Time

time time ing time time time

delivery accu-

physical shipping ac-

racy; on-time

storage inventory picking ac- curacy; or-

Quality delivery;

accuracy accuracy; curacy ders shipped

cargo damage

stock-out rate on time

rate

distribution

Cost inventory cost

cost

inventory

receiving picking pro- shipping transport uti-

Productivity space utiliza-

productivity ductivity productivity lization

tion; turnover

Dimensions Process - Transversal Indicators

Inbound Processes Outbound Processes

Dock to stock time Order lead time

Time

Global= Queuing time

Order ll rate, Perfect orders

Quality

Global= Customer satisfaction, Scrap rate

Order processing cost

Cost

Global= Cost as a % of sales

Outbound space utilization

Productivity

Global= Throughput

2.6. Classication of the Warehouse Performance Indicators 33

In Table 2.12 we propose a mapping for both the specic (on the upper half of the table) and

transversal indicators (on the lower half ) over the warehouse activities. The activities are

given on the columns. Although inventory is not a warehouse activity, we choose to include

it in Table 2.12 due to its importance in warehouse management. Gallmann and Belvedere

(2011) state that companies consider inventory management as a key to achieve excellent

service levels. We also observe numerous metrics treating the subject (see Section 2.4).

On the rows of Table 2.12, it is possible to observe the previous classication dimensions

(time, quality, cost and productivity). Each direct indicator is then placed in the related

cell in the table. For example, order picking time is a time indicator which is specic to

In the lower half of Table 2.12, we illustrate the direct transversal indicators. Chan and

Qi (2003) have dened that the inbound logistics concern both the materials transportation

and storage, while outbound logistics involve the outbound warehousing tasks, transporta-

tion and distribution. Based on this idea, the inbound process covers both Receiving

and Storage activities and are named as Inbound Processes in Table 2.12 while Picking,

Shipping and Delivery activities are regrouped under Outbound Processes. Inventory is

considered as internal process in this case linking inbound to outbound processes. The

indicators are then placed according to the extent of their boundaries. For example, the

ing receiving and storing activities. Order lead time is an outbound indicator, covering

picking, shipping and delivery activities. Moreover, there are the global transversal indica-

tors that cannot be assigned to specic activities. That is the case, for example, of Cost

as a % of sales, dened as global to all warehouse activities since its measure represents

a sum of warehouse activity eorts. Second, the throughput indicator was classied as

a global measure inside the warehouse, since it assesses the quantity of products that are

produced by the warehouse in items per hour (Voss et al., 2005), not including the delivery.

We note that the boundaries of indicators as described in Table 2.12 depend on ware-

the boundaries of the indicators dierently. The operating strategies impact mainly the

transversal indicators. One example is the order lead time. If a make-to-order system is

considered, the customer order would start upstream (in the supply process), not at the

picking activity.

Some remarks can be made on Table 2.12 based on the shown empty elds. First of

all, it is important to note that the empty cells in Table 2.12 do not mean that there is

no paper analyzed has used an indicator related to the activity/process. In Table 2.12, it

could be seen that the receiving and storage activities are less covered than the outbound

areas. This shows that the statement of Gu et al. (2007) that the research on receiving

is limited, is still valid. The number of outbound indicators is higher than the number

of indicators for the inbound processes. This is not very surprising as the warehouse

activities are getting more and more customer oriented. So, it is possible to conclude that

the outbound processes are considered as more critical than the inbound ones and hence

34 2. Literature Review on Warehouse Performance

are subject to more control. The same discussion is also true for the inventory.

Some indicators are directly related to resources used in the warehouse. Such indicators

impact all warehouse activities. Therefore, instead of presenting them in Table 2.12, we

choose to classify them as resource related indicators. There are 2 major resources: labor

and equipment. The facilities are considered in the same group as equipment. The related

Dimensions

Resource Related Indicators

Labor Equipment and Building

Time Equipment downtime

Quality

Cost Labor cost Maintenance cost

Productivity Labor productivity Warehouse utilization

Analyzing Table 2.13 we note some empty cells for time and quality dimensions. The

rst empty cell is labor time, which is usually utilized as a data instead of an indicator.

The labor time is used to measure several productivity indicators, thus, it is not utilized

in warehouses for performance indicator purposes. For the cell quality versus labor, it is

expected because the quality of work is usually measured for each activity separately (e.g,

accuracy in picking, shipping; see Table 2.12) instead of a general way. The cell equipment

2.7 Conclusions

Some conclusions can be made from the reported results.

In general, the works diversify a lot in terms of performance area evaluated and the mea-

surement tool used for it. The warehouse area means the evaluation of one/various types of

warehouse with focus on one/several warehouse activities. The papers' results are usually

very specic for one kind of situation. For example, works related to tobacco industry

warehouse (Wang et al., 2010), a DC of fresh products (Manikas and Terry, 2010) or an

air force warehouse (Sohn et al., 2007) have used dierent mathematical tools and indica-

tors to evaluate performance. Other dierences are in the type of warehouse studied (e.g.

According to Section 2.3.2 the majority of our database has performed surveys to treat

warehouse performance subject. This shows a new tendency of studies in two directions:

to nd relationships among dierent warehouse performance areas (e.g. degree of automa-

tion inuencing warehouse productivity (De Koster and Balk, 2008)); and the evaluation

of concepts not usually expressed as ratios and, therefore, not measured yet (e.g. VAL

2.7. Conclusions 35

From these papers, it can be concluded that a high degree of automation has a positive

impact on the delivery accuracy and the total cost (incl. depreciation and maintenance).

This result was expected; otherwise, it would be more ecient to work with people and low

automation. About the use of metrics to manage the information systems (WMS, RFID)

we could see that they are not applied in warehouses. The indicators about information

systems are usually designated to evaluate systems on the implementation phase (based on

time/resources savings). After that, the managers generally use the indexes provided by the

system to evaluate all other warehouse areas. For VAL activities, the studies evaluate their

growing importance in warehouse operations and determine the low value and high-end

activities.

The human resources management in warehouses is an area that has attracted increas-

ing attention in the literature. Several papers of our database treat the operational labor

tivity goals and customer satisfaction. One reason for the importance of this subject is

reported by Park (2008) who highlights that the front-line distribution center personnel

could be responsible for any task in moving products inside the distribution center. Any

service failure or inecient performance directly increases customer order cycle time and

negatively impacts the level of service as perceived by the customers. It can be seen from

papers' application area that the majority of researches have considered manufacturing

companies, which usually employ people to execute the warehouse activities since automa-

tion is a high investment for enterprises that do not have their focus on logistics.

Even if there is a tendency for indirect measures", they are not used for daily manage-

ment since they require a great quantity of data sometimes dicult to obtain. So, direct

The total direct indicators sum 38 measures, from which 9 of time, 13 of quality, 6

of cost and 10 of productivity. There are indicators related to one activity/area (e.g.

shipping productivity) or several (e.g. dock to stock time, cost as a % of sales). Analyzing

the application area of indicators (i.e. the activity measured by the indicator) we can

conclude that half of them are related to outbound activities (i.e. picking, shipping and

delivery). This reveals that the outbound processes/activities are considered more critical

than the inbound ones and hence they are subjected to more control.

cators. In this framework we classify indicators not only according to quality, cost, time

and productivity dimensions, but also in terms of warehouse activities (receiving, storage,

picking, shipping and delivery). The result of this classication shows that the number of

outbound indicators is much higher than the number of inbound indicators. This is not

very surprising as the warehouse activities are getting more and more customer oriented.

An important evidence we can highlight is that literature about the performance anal-

ysis and management of the Distribution Center (DC) operations is not as abundant as

for the location and cooperation problems (Dotoli et al., 2009). Indeed, we have not found

literature reviews focusing specically on warehouse performance management and its in-

dicators.

The low attention given for warehouse performance subject leaves several gaps that

36 2. Literature Review on Warehouse Performance

warehouses. Only some works evaluating the inuence of indicators on the warehouse per-

formance (e.g. Voss et al. (2005); De Koster and Balk (2008)) can be reported. The next

chapter details these works regarding indicator or process relationships since this disserta-

tion also measures indicator relations. Additionally, works about performance aggregation

are presented to verify the main developments made in this theme and the mathematical

Chapter 3

Literature on Performance Integration and Tools

audacieuse.

Isaac Newton

Contents

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3.2.1 Literature on indicator relationships and indicators aggregation . . . . . 38

3.2.2 Literature on Performance Integration . . . . . . . . . . . . . . . . . . . . 40

3.3 Overview on mathematical tools used for performance inte-

gration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

3.3.1 The choice of the dimension-reduction statistical tool . . . . . . . . . . . 43

3.3.2 Principal Component Analysis - PCA . . . . . . . . . . . . . . . . . . . . 44

3.3.3 Factor Analysis - FA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

3.3.4 Canonical correlation analysis - CCA . . . . . . . . . . . . . . . . . . . . 50

3.3.5 Structural Equation Modeling - SEM . . . . . . . . . . . . . . . . . . . . 52

3.3.6 Dynamic Factor Analysis - DFA . . . . . . . . . . . . . . . . . . . . . . . 54

3.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

Abstract

We rst present the results of the literature review about indicator relationships and

performance integration. The gaps are identied as well as the mathematical tools

used to associate indicators. The general characteristics of the main techniques are

presented to provide theoretical basis for the methodology development.

38 3. Literature on Performance Integration and Tools

3.1 Introduction

The objective of this Chapter is twofold: to describe works related to indicator relationships

and/or performance integration and to present the mathematical tools used in these papers.

To reach the rst objective, a non exhaustive literature search is performed on online

databases. The keywords used are related to performance integration, performance ag-

publications (journal articles, conference proceedings, etc.) are included in the database

search. Due to the great variation of objectives and applications in the papers found, we

only present the most related work to this dissertation in the next sections.

The second goal of this chapter is to present the mathematical tools used in the earlier

works to relate indicators or aggregate performance measures. From the articles analyzed,

it is possible to identify some groups of tools utilized with distinct objectives. Therefore,

a general presentation of these groups is made, with a special attention on the statistical

tools used for dimension reduction, which allow the indicators aggregation.

3.2.1 Literature on indicator relationships and indicators aggregation

Papers that dene indicator relationships are not new. It is possible to identify two main

development periods on this theme. First, the papers try to identify if there are indi-

cator relationships; then, these relationships are measured. This measurement is made

qualitatively (using decision making tools such as AHP (Analytic Hierarchy Process)) or

quantitatively. An example of the rst period is the work of Bititci (1995), which uses

a QFD matrix (Quality Function Deployment) to display how measures of dierent lev-

els (strategical, tactical, operational) are inuencing each other according to manager's

perception. In the same work the author models the process for each strategic measure

performance results.

In the relationships measurement period, the work of Suwignjo et al. (2000) develops

the Quantitative Model for Performance Measurement System (QMPMS) to quantify the

eects of factors on performance through the AHP utilization, which is based on manager's

opinion. The three main steps of QMPMS are: (i) identifying factors that aect perfor-

mance and their relationships, (ii) structuring the factors hierarchically, (iii) quantifying

the eect of the factors on performance. The authors discuss that even the methodology

seems intuitive; one of the problems to measure relationships quantitatively is the quali-

tative nature of some measures, for example, management commitment (Suwignjo et al.,

2000).

management literature some years later. This approach is the statistical techniques of

measures. For example, the work of Fugate et al. (2010) investigates the inuence of lo-

3.2. Literature Review 39

Logistics Organizational

Performance Performance

Effectiveness Efficiency Differentiation

Figure 3.1: Relationships among logistics variables. Source: Fugate et al. (2010)

and dierentiation and the authors assume that all three are related. A questionnaire is

performed with industry managers to obtain the necessary database for the tool applica-

tion. At the end, the results suggest that the overall performance of the logistics function

should produce high levels of logistics eectiveness, eciency, and dierentiation, aecting

Another example is Cai et al. (2009), proposing a framework to analyze and to select

the right key performance indicators (KPI) to improve supply chain performance. The

framework assigns priorities to dierent KPIs and uses PCTM (KPI cost transformation

matrix) to verify the cost incurred for the KPI accomplishment, considering also the extra

cost caused in all other dependent KPIs. The authors interview managers and employees

identifying 20 dierent KPIs and dening their coupled relationships. Then, the cost of

each KPI accomplishment with its relationships is estimated from interviews with man-

agers. The relationships between two dependent KPIs accomplishment costs are measured

quantitatively by the following classication: weak (0,05), neutral (0,25), and strong (0,5).

Coskun and Bayyurt (2008) determine the eects of the indicator measurement fre-

Measures, Customer Measures, Process Measures, Learning and Growth Measures). The

relations between the measurement frequency of performance indicators and the corporate

human judgment. That is the case of Rodriguez et al. (2009) proposing a methodology to

identify KPI relationships and projecting them on strategic objectives, to know whether

the upstream objectives are being reached or not. Principal Component Analysis (PCA) is

performed to quantify indicator relationships and group them according to these relations.

outlined.

Patel et al. (2008) develop a methodology to demonstrate the cause and eect rela-

tionships between the components of the performance rating system. Using Structural

Equation Modeling, a causal-loop diagram showing the cause and eect relationships be-

tween the 16 common performance indicators is constructed based on a data set of two

40 3. Literature on Performance Integration and Tools

years. These relationships are used to draw scenarios regarding an organization's future

performance.

Johnson et al. (2010) identify the operational policies, design characteristics, and at-

tributes of warehouses that are correlated with greater technical eciency, i.e. those factors

that impact warehouse performance. The variables correlated with high eciency are iden-

tied using a regression model and solve it using ordinary least squares. Another work using

regression model to assess performance is by Kassali and Idowu (2007), which denes the

factors determining the operational eciency of onion storage and uses statistical inference

sications. Bititci (1995) denes that indicators may have simple or complex relationships;

in other words, if one indicator changes this may alter one or more data items elsewhere

in the information system. Suwignjo et al. (2000) improve the classication of indica-

tor relations as direct (vertical) eect (an indicator inuences another of a higher level),

indirect (horizontal) eect (an indicator inuences another indicator of the same level),

self-interaction eect (the indicator inuences itself ). Cai et al. (2009) classify the rela-

tionships into three categories: parallel, sequential and coupled. In a parallel relationship,

two KPIs are independent of each other, i.e. the eorts of accomplishing these two KPIs

are not related. A sequential relationship usually implies a simple cause-eect relationship,

but the reverse dependence does not always hold. Finally, the coupled relationship means

To the best of our knowledge, the term performance integration is interpreted in two dif-

indicator system framework which links the measures to strategy. One such example, as

dierent aspects of an organization's performance (e.g. the Tableau de Bord ) that feeds

the three levels of management (strategy, management and operations). This aggregation

usually deals with the translation of all the elementary performance expressions associated

with the various heterogeneous criteria into a common reference (cost or degree of satisfac-

tion) (Clivillé et al., 2007). In these works, usually the number of indicators from higher

levels is reduced to allow managers to control just the key parameters, i.e. the key perfor-

mance indicators. The literature about this kind of performance integration is signicant,

with several methods proposing the establishment of a performance indicator group (e.g.

SCOR model - Supply Chain Operations Reference-model) or dening how the indicators

The second kind of performance integration, which is studied in this dissertation, refers

to the performance measurement in a global view, not excluding indicators but aggregating

them to nd out the total performance of an area or enterprise. Franceschini et al. (2008)

refer to the performance integration as the association of informations from one or more

sub-indicators in just one aggregated and synthesized indicator. The number of papers

literature when compared to the rst interpretation. The following papers are related to

3.2. Literature Review 41

Chan and Qi (2003) develop a process-based model to measure the holistic perfor-

mance of complex supply chains. They consider productivity, eciency and utilization as

composite measures since they relate inputs and outputs. A group representing various

management areas of the supply chain is formed and the expert opinions are incorporated

Lohman et al. (2004) present a prototype system that basically is a balanced scorecard

tailored to the needs of the company studied. After the performance indicator system

metric has a dierent dimension, the authors suggest a method for normalizing metrics

linearly.

In Sohn et al. (2007), the authors developed an Air Force Warehouse Logistics Index

(WLI) to evaluate the logistics support capability of ROKAF (Republic of Korea Air Force)

warehouses. Even if the main goal is not performance measurement, the constructed index

takes into account relationships among various inuential factors for warehouse capability.

The dataset is obtained by interviews with warehouse employees and the answers are related

to latent variables using Structural Equation Modeling (SEM). The six latent variables

sj , with j = 1...6 inuence WLI, which contributes to logistics support capability and

warehouse modernization. The relationship between the overall logistics index ηi and the

six observed variables yij , with i referring to each respondent is (Equation 3.1):

Luo et al. (2010) propose a hierarchical model of performance factors to assess the

(Fuzzy Analytic Hierarchy Process) is used to calculate index weight, then fuzzy compre-

The work of Jiang et al. (2009) develops a theoretical indicator system of logistics perfor-

mance with the objective to analyze the interactions among these performance measures

and to optimize them. The dimensions of logistics performance measurement are time,

quality, cost, exibility (see Figure 3.2) and each dimension includes several indicators.

Service

Quality

Logistics

Costs

Logistics

Flexibility

Figure 3.2: Framework to evaluate logistics performance in supply chains. Source: Jiang

et al. (2009)

42 3. Literature on Performance Integration and Tools

is the utilized tool to optimize the index system and delete the indexes with small relational

grade. Finally, DEMATEL is also applied to evaluate the weight of each index and the

surate elementary performances and the relative weights of the performance measures from

et al. (2007) use MACBETH with Choquet integral operators to take into account the

excellence for products, services, or processes, and then making the improvements necessary

to reach those standards, commonly called best practices (De Koster and Balk, 2008).

(Gu et al., 2010). In these cases, DEA (Data Envelopment Analysis) is probably the most

widely used mathematical approach for benchmarking of organizational units (Jha et al.,

2008).

Data Envelopment Analysis (DEA) is regarded as an appropriate tool for this task

because of its capability to capture simultaneously all the relevant inputs (resources) and

outputs (performances) using one single performance factor, to construct the best perfor-

mance frontier, and to reveal the relative shortcomings of inecient warehouses (Gu et al.,

2010).

Some examples of this kind of works are by Schefczyk (1993), Ross and Droge (2002)

and Johnson et al. (2010). The recent work of Andreji¢ et al. (2013) proposes to benchmark

DCs using PCA (Principal Component Analysis) before the DEA. The PCA is applied for

inputs and outputs separately to reduce the number of variables for the DEA model.

literature review. First, the majority of works develop a methodology for performance

aggregation using statistical tools; however, the indicators aggregation is not included as

a step before attaining the global performance. Only the work of Jiang et al. (2009)

achieves global performance through indicator relationships. However, these relations are

dened based on expert judgments. This situation demonstrates the second characteristic:

the works proposing aggregated indicators to represent the global performance usually

utilize methods based on expert judgments. One exception is the work of Rodriguez et al.

(2009), which has already aggregated performance indicators in factors without human

judgment. However, these factors are not yet transformed in a global performance. Hence,

this dissertation comes to fulll this gap, providing a global warehouse performance through

In the next sections, we present an overview on the mathematical tools used in the most

relevant papers. A special attention is given to statistical tools which allow performance

indicators' aggregation.

3.3. Overview on mathematical tools used for performance integration 43

integration

The goal of this section is twofold: (i) to identify the most appropriate mathematical tools

to attain indicators aggregation without human judgment; (ii) to provide a basic overview

of these chosen mathematical tools, focusing on the requirements for their application and

From the papers presented in the above section, we note that dierent kinds of mathe-

matical tools are used to assess performance. It is possible to divide the tools in dierent

There is a vast literature and numerous tools to help decision makers. Several papers

treat the relationship among indicators using decision support systems. The majority of

these tools interpret the manager's opinion about indicator relationships and weights in a

methods as AHP is that they have judgments as inputs, which can be incongruent with

nd out relationships from the indicator equations and their data collected periodically,

without manager judgment. Thus, methods which incorporate manager's opinions like

AHP, FAHP (Fuzzy Analytic Hierarchy Process), DEMATEL (Decision-Making Trial and

Based Evaluation TecHnique) and Fuzzy are not considered in our analysis.

of dierent DMUs (Decision Making Units), based on multiple inputs and outputs. In

DEA approach, essential input and output data are selected and the set of observed data

is used to approximate the Production Possibility Set (PPS). The PPS represents all input

and output combinations that actually can be achieved. The boundary of the PPS is called

the ecient frontier and characterizes how the most ecient warehouses trade o inputs

and outputs (Johnson et al., 2010). The eciency is relative and relates to the set of units

within the analysis, i.e. the warehouses are ecient among the other units (Andreji¢ et al.,

2013).

Even if it is possible to use DEA to analyze just one DMU over time (another application

besides the benchmarking), it does not satisfy our objectives in some aspects. Firstly, we

want to dene the indicator relationships to provide the managers additional information

about the impacts of the decisions that are going to be taken based on performance results.

DEA does not give information about input and output relationships. Secondly, the dataset

(inputs and outputs of the model) used for eciency analysis are operational data, and not

the indicator results as we intend to use in this work. Therefore, DEA is also not utilized

in this dissertation.

Looking at the statistical literature, the multivariate analysis has got the potential

to identify relationships between variables over time, clustering them according to these

relationships. Additionally, these tools can aggregate variables determining their weights

and reducing the dimension of the analysis to help managers in decision-making situations.

44 3. Literature on Performance Integration and Tools

Some techniques presented in next sections are: Principal Component Analysis, Factor

Factor Analysis. Among these tools, only Dynamic Factor Analysis is specially designed

for time series data, whereas the others have better results with other kinds of data. As

an example, Hoyle (2012) cites that standard SEM approaches use variables measured on

categorical data (e.g. true-false). However, the use of these tools with time series data is

not forbidden, but in some cases adaptations need to be made for their application.

As these dimension-reduction tools are associated with this dissertation's proposal, they

3.3.2.1 Objective

Principal Component Analysis (PCA) is one of the most common types of multivariate

methods to identify association patterns between variables (Katchova, 2013). A PCA of-

ten uncovers unsuspected relationships, allowing you to interpret the data in a new way

(Minitab Inc., 2009). The main purpose is to reduce the information of many observed vari-

ables into a little group of articial variables named components (Manly, 2004). In PCA,

the components empirically aggregate the variables without a presumed theory (Wainer,

2010).

There is no specicity about the kind of data that should be used to perform PCA. The

specially when PCA is used for data reduction or exploratory purposes. However, some

authors suggest that the PCA can provide better results if data follow a normal distribution.

The sample (dataset) is a matrix n×p with n number of observations for each p variable.

Usually, the inputs come from questionnaires (each observation is a dierent person), but

nothing prohibits the use of other types of data as, for example, time series.

There are some conditions that the dataset should satisfy (Manly, 2004):

If the number of variables is greater than the number of observations, as some practical

cases within the performance management context, the application of classic PCA presents

problems. The solution could be to apply the NIPALS (Nonlinear Iterative Partial Least

et al., 2010).

Besides the sample size, PCA is sensitive to great numerical dierences among variables.

Therefore, after the acquisition of the minimum number of observations required, it is often

detailed in Chapter 4.

3.3. Overview on mathematical tools used for performance integration 45

The principal components are dened in order to capture the greatest variance of the

dataset. They are calculated by nding the variable eigenvalues and eigenvectors of the

covariance matrix for the p variables. The eigenvalues are a numeric estimation of the

variable variation explained by each component (Wainer, 2010). In the case of PCA, all

variance of the observed variables is analyzed (shared, unique and error variances) (Manly,

2004). Moreover, PCA considers that the variables comprise only linear relationships.

The PCA method essentially denes the same number of components as the quantity of

plot. As explained in the sequence, the number of components explaining the total dataset

variance can be less than the total number of variables depending on the data character-

istics.

Let us consider that Figure 3.3 demonstrates the scatter plot of indicators measured

monthly. The X axis represents the time and Y axis the indicator values. The points in

the graphic are the observations (indicator values) in all periods of time. In Figure 3.3, the

rst and second principal components are u and v, representing the rst and the second

greatest variance of the dataset, respectively. The u and v components are orthogonal

demonstrating that they are uncorrelated to each other. It happens to all components

(Wainer, 2010).

Figure 3.3: Scatter plot of the dataset with the rst and second principal components.

From the p variables X1 , X2 , . . . , Xp , each principal components C1 , C2 , . . . , Cp describes

a dimension of data variation (Manly, 2004). Since each component is a linear combi-

nation of the observed variables, the principal components (Ci ), combining the variables

p X

X p

Ci = aij × Xj (3.2)

i=1 j=1

The outcome of PCA is principal components like Equation 3.2, since the maximum

46 3. Literature on Performance Integration and Tools

number of components extracted always equals the number of variables (Minitab Inc.,

2009).

It is important to note, in Equation 3.2, that not all variables are included in every prin-

cipal component. Just the original variables that account for the data variance explained

Principal components resulted from PCA are ranked in a descending order of impor-

tance, such that V ar(C1 ) ≥ V ar(C2 ) ≥ . . . ≥ V ar(Cp ), where V ar(Ci ) denotes the vari-

The aij in Equation 3.2 are the coecients of the variables with i = 1, 2, . . . , p and

j = 1, 2, . . . , p. These coecients mean the correlation between the original variables and

the component. It could be also interpreted as the relative weight of each variable in the

component Ci . Thus, the bigger the absolute value of the coecient, the more important

the corresponding variable is in constructing the component (Minitab Inc., 2009). These

loadings (notation used in this thesis) are optimally dened in PCA analysis to produce

the best set of components which explain the maximum variation of the observed variables.

The loadings have the constraint presented in Equation 3.3. The squared loadings

p X

X p

a2ij = 1 and aij ∈ < (3.3)

i=1 j=1

• Enter data (in the form of covariance or correlation data matrix) in a software which

• Interpretation of results.

An important part of PCA is the interpretation of the results and its main task is to

determine the number of principal components that will be retained to represent data.

between simplicity (retaining as few as possible) and completeness (explaining most of the

data variation) (Katchova, 2013). Usually, the rst few principal components are chosen

One of the PCA objectives is to explain the maximum amount of variables variance in

this component. However, the results are not always easily interpretable. To help with

this decision, there is the Kaiser's criterion. Kaiser's rule determines that principal com-

ponents with eigenvalues bigger than 1 (λ > 1) should be retained. The eigenvalues of the

correlation matrix are equal to the variances of the principal components, thus, eigenvalues

3.3. Overview on mathematical tools used for performance integration 47

The scree plot can also help in PCA interpretation. This graphic shows the variance

of the data (y axis) explained by each component (x axis) (see Figure 3.4 for an example).

The principal components are sorted in decreasing order of variance, so the most important

principal component is always listed rst. The objective is to help analysts to visualize

the relative importance of the components, identifying easily the sharp drop in the plot

variance.

It is also possible to decide on the number of principal components based on the amount

of explained variance. For example, you may retain components that cumulatively explain

Finally, the decision on the number of principal components retained can be based on

Even if the presented techniques provide a useful basis to choose the number of com-

ponents, the analyst should know that all components must be interpretable (Rodriguez-

Rodriguez et al., 2010). Since the components are synthetic variables which do not have a

specic unit of measurement, it is important to nd their meaning in the analysis carried

out.

3.3.3.1 Objective

Factor Analysis (FA) is widely used to analyze data because users nd the results useful for

gaining insight into the structure of multivariate data (Manly, 2004). Factor analysis has

aims that are similar to those of Principal Component Analysis, i.e. describe data in a far

both Factor Analysis and Principal Component Analysis summarize variables considering

The main dierence between PCA and FA is that PCA is not based on any particular

statistical model whereas FA is based on a model (Manly, 2004). It means that Factor

Analysis assumes the existence of a few common factors driving the data variation and

48 3. Literature on Performance Integration and Tools

Principal Component Analysis does not make such assumptions (Katchova, 2013). More-

over, PCA uses all types of variance to estimate components whereas FA utilizes only the

There are two most common factor analysis methods: EFA (Exploratory Factor Anal-

ysis) and CFA (Conrmatory Factor Analysis). The EFA is used to search possible under-

lying structures in the variables while CFA's goal is to conrm with the data a predened

An extension of the FA is the multiple factor analysis, which analyzes several data

tables at the same time. These tables measure sets of variables collected on the same

observation or the same variables are measured on dierent set of observations (for details,

There are some conditions to perform CFA:

• sample bigger than 150 observations for each variable, or the sample size should have

The last condition limits the utilization of time series as inputs in the model.

The FA model postulates that the observable random variable vector X (with p observa-

tions) is linearly dependent upon a few unobservable random factors F1 , F2 . . . , Fm and p

additional sources of variation ε1 , ε2 . . . , εp called errors or specic factors (Johnson and

Wichern, 2002).

The dimensions (or factors) are formed by the combination of observed variables highly

correlated. The objective is to identify the latent dimensions contained in data; i.e. to

group the variables in dimensions that represent them. The explanation degree of each

The representation of the factors is given by Equation 3.4 (Johnson and Wichern, 2002).

where Fj is the common factor and j = 1, 2, . . . , m; bij are the factor loadings of the

i th variable on the j th factor; Xi are the variables with i = 1, 2, . . . , p; εi is the variation

of Xi that is not explained by the factors Fj .

The factor loadings are measured by the FA model, representing how much a factor

explains a variable. High loadings (positive or negative) indicate that the factor strongly

3.3. Overview on mathematical tools used for performance integration 49

inuences the variable whereas low loadings (positive or negative) indicate a weak inuence.

It is necessary to examine the loading pattern to determine on which factor each variable

loads. Some variables may load on multiple factors (Minitab Inc., 2009).

in X attributable to the common factors (Katchova, 2013), i.e. it assesses the quality of

the measurement model for each variable (Krizman and Ogorelc, 2010). Communality is

measured by the sum of squares of the loadings of the i th variable on the m common

The higher the communality value, the more the variable is explained by common

factors. This parameter is also used in the analysis of FA results as well as loadings. For

example, Krizman and Ogorelc (2010) dene in their paper that variables with a loading

of less than 0.75 and communality less than 0.40 were discarded.

Some steps to perform Factor Analysis are, according to Costa (2006), as follows:

• Perform rst a PCA and verify the number of factors that should be used (analyzing

what kind of data each factor represent), when one does not know the variables'

behavior.

• Rotation of factor loading. This procedure (using e.g. Varimax rotation) rotates the

factor to get the higher number of factor loadings as possible. It helps to interpret

A common tool used to provide visual information about the factors is the scree, or eigen-

value, plot (graph of factors versus the corresponding eigenvalues). From this plot, you

can determine how well the chosen number of components t the data.

ucts, enterprises) the factor analysis can be measured separately for each group; it can

avoid the designation of variables from dierent natures in the same factor.

Finally, the diculty to interpret the variable clusters of the unrotated factor loadings

can be overcome with their rotation, which simplies the loading structure, allowing the

analyst to more easily interpret the results. The goal of factors rotation is to nd clusters

of variables that, to a large extent, dene only one factor (Katchova, 2013).

There are two kinds of rotation: orthogonal and oblique. The orthogonal rotation

preserves the perpendicularity of the axes (rotated factors remain uncorrelated). The

oblique rotation allows the correlation between the rotated factors, and the main method is

axes leading to new axes that are not perpendicular (Johnson and Wichern, 2002).

There are four methods to orthogonally rotate the initial factor loadings (Minitab Inc.,

2009):

50 3. Literature on Performance Integration and Tools

• Equimax - maximizes variance of squared loadings within both variables and factors.

• Varimax - maximizes variance of squared loadings within factors (i.e. simplies the

columns of the loading matrix); the most widely used rotation method. This method

• Orthomax - rotation that comprises the above three depending on the value of the

Nevertheless, Johnson and Wichern (2002) arm that the choice of the type of rotation

is a less crucial decision. For them, the most satisfactory factor analysis are those in which

rotations are tried with more than one method and all the results substantially conrm

3.3.4.1 Objective

Canonical Correlation Analysis (CCA) is a method for exploring the relationships between

two multivariate sets of variables. CCA is similar to multiple regression in assessing variable

relationships. The main dierence is that multiple regression allows only a single dependent

variable whereas CCA analyzes multidimensional relations between multiple dependent and

independent variables (Coskun and Bayyurt, 2008). Therefore, CCA has, as main objective,

to measure the relationships within each variable set, independent and dependent, and also

between both (Voss et al., 2005). For the purposes of this thesis, we are interested in the

Canonical correlation analysis is not recommended for small samples. Moreover, multi-

variate normal distribution assumptions are required for both sets of variables (UCLA,

2012). Unlike Principal Components Analysis, standardizing the data has no impact on

The aim of CCA is to nd a linear combination of the independent (or predictor) variables

such that the outcomes has the maximum correlation with the dependent (or criterion)

To demonstrate how this result is attained, let us consider two set of variables X and

the linear combinations of the rst set of variables, X (Equation 3.6), and V corresponds

to the second set of variables, Y (Equation 3.7) (PennState, 2015b). For computational

3.3. Overview on mathematical tools used for performance integration 51

U2 =a21 × X1 + a22 × X2 + . . . + a2p × Xp

.

. (3.6)

.

V2 =b21 × Y1 + b22 × Y2 + . . . + b2q × Yq

.

. (3.7)

.

Each member of U will be paired with a member of V, forming the canonical variates.

Canonical dimensions, also known as canonical variates, are latent variables that are analo-

gous to factors obtained in factor analysis. In general, the number of canonical dimensions

is equal to the number of variables in the smaller set; however, the number of signicant

For example, (U1 , V1 ) is the rst canonical variate and the objective is to nd the

coecients (ai1 , ai2 , . . . , aip and bi1 , bi2 , . . . , biq ) of the linear combinations that maximize

the correlations between the members of each canonical variate pair (PennState, 2015b).

The canonical correlation (Rc ) for the ith canonical variate pair is given by the covariance

(cov ) of the canonical variate pair per the square root of variances (var ) of Ui and Vi

(Equation 3.8):

cov(Ui , Vi )

Rc = p (3.8)

var(Ui )var(Vi )

The output of canonical correlation consists of two parts, canonical functions and canonical

variates. Each canonical function is composed of two canonical variates, one independent

and one dependent. The independent and the dependent canonical variates represent, each

one, the optimal, linear and weighted combination of the variables that correlate highly

The correlation between the independent and dependent variates in each function is

assessed by the canonical correlation coecient (Rc ) and the shared variance between

2 Multiple

canonical functions are then derived that maximize the correlation between the independent

and dependent canonical variates, such that each function is orthogonal to all others (Voss

et al., 2005).

52 3. Literature on Performance Integration and Tools

Two main analytical ndings can be secured from CCA results: (i) the evaluation of how

many dimensions (canonical variables) are necessary to understand the association between

the two sets of variables; (ii) to explore the associations among dimensions and how much

To interpret each component, we must compute the coecients (also named loadings)

between each observed variable and the corresponding canonical variate (UCLA, 2012).

The magnitudes of the loadings give the contributions of the individual variables to the

3.3.5.1 Objective

Structural Equation Modeling (SEM) is a growing family of statistical methods for modeling

the relations between variables. The method is also known as Covariance Structural Equa-

SEM is appropriate for complex, multivariate data and testing hypotheses regarding

relationships among observed and latent variables, the two broad classes of variables in

To perform SEM, it is necessary to be aware that the sample size and number of parameters

to be estimated can make SEM unadvisable. Several estimation issues arise in SEM when

N and some alternatives have been developed to handle this kind of data (Chow et al.,

2010). There is no rm decision rule for the minimum sample size for SEM, but several

authors suggest that at lower sample sizes, typically below 150, structural models with

latent variables become unreliable. Furthermore, there are similar advices against the use

of SEM in cases where the ratio of sample size to estimated parameters is less than 10

(Autry et al., 2005). In cases where there is a relatively small sample size, the threshold

values for factor loadings and communalities are, sometimes, increased, and Partial Least

Squares Regression (PLS) is usually employed to assess the measurement model (Krizman

Some dataset requirements to apply SEM are (Bentler and Chou, 1987):

• Simple random sampling - each of the units or cases have the same probability to be

3.3. Overview on mathematical tools used for performance integration 53

series data raises the issue of autocorrelated errors. There are methods for accommodating

autocorrelated errors in structural equation models, but they are complex and will not

SEM comprises the ability to construct latent variables: variables which are not measured

directly, but are estimated in the model from several measured variables. SEM requires

a theoretical model specication before its application. Thus, as the Conrmatory Factor

Analysis (CFA), an accurate estimation of the latent variables depends on the quality of the

theoretical model constructed. The test of the structural model constitutes a conrmatory

assessment of the hypothesized causal relationships among the constructs (Krizman and

Ogorelc, 2010).

The theoretical model of SEM can have numerous congurations. Initially, the model

can have variables that are dependent and independent in the same model. For instance, a

set of observed variables might be used to predict a pair of constructs (or latent variable)

that are correlated, uncorrelated, or related in such a way that one forms the other. In the

latter case, one of the dependent variables is also an independent variable since it is used

Another conguration of theoretical models regards the construct specication (i.e. the

aggregation method used to dene the latent variables) which can be classied as reective

of a weighted sum of variables, i.e. the measured variables cause the construct. In the

Figure 3.5 shows the reective construct model represented by the path diagram, which

is a graphical representation of direct and indirect eects of observed and latent variables.

In this model, Y and X are the latent variables operationally dened by the measured

54 3. Literature on Performance Integration and Tools

The path analysis, essentially:(i) helps in the understanding of correlations patterns among

the variables; (ii) explains as much of the variable variation as possible with the model

specied. In summary, after the denition of the theoretical model, it is tested using the

dataset of the observed variables and the results will infer about whole hypothesized model:

There are two general causal modeling approaches to model measurement: the covariance-

based method and the partial least squares (PLS). Covariance-based methods are more ap-

propriate for conrming theory and parameter estimation, and require large samples sizes

with normal distribution. PLS, in contrast, is more appropriate when theory is lacking

regarding the nature of relationships among constructs, dimensions and their indicators

The interpretation of path coecients cannot be done straightforward (Kline, 2011). The

higher the correlation among multiple indicators of a given construct, the more consistent

i.e., reliable, the measures. However, they are not correlation coecients. Suppose we have

a network with a path connecting from region A to region B. The meaning of the path

coecient theta (e.g., -0.16) is this: if region A increases by one standard deviation from

its mean, region B would be expected to decrease by 0.16 its own standard deviations from

its own mean while holding all other relevant regional connections constant.

3.3.6.1 Objective

Dynamic factor analysis is a special case of MARSS Model (Multivariate Autoregressive

State Space Model). State-Space modeling techniques are originally developed as single-

subject time series estimation tools (Chow et al., 2010), studying linear stochastic dynamics

DFA can be looked at as a super regression model especially designed for time-series

examining correlates of a single summary metric (i.e. an output), DFA can provide infor-

mation on correlation (explanatory variables) of patterns that emerge over time (Hasson

and Heernan, 2011). Thus, DFA explains temporal variation of a set of n observed time

series (variables) using linear combinations of m hidden trends (or common trends), where

m << n (Holmes et al., 2014).

Although DFA has potential as a useful analysis technique, it often takes an unusually

long time to converge (often exceeds several hours as larger the dataset and the number of

common trends). The results also tend to become inconsistent with such large data sets

(Holmes et al., 2014). Therefore, DFA brings good results when n (number of observed

3.3. Overview on mathematical tools used for performance integration 55

Besides the short dataset, DFA also accepts non-stationary time series with missing

Dynamic Factor analysis manages to combine, from a descriptive point of view (not proba-

bilistic), the cross-section analysis through Principal Component Analysis (PCA) and the

time series dimension of data through linear regression model (Federici and Mazzitelli,

2005). DFA models observations in terms of a trend, seasonal eects, a cycle, explanatory

A limitation of DFA is that the common trends are combined in a linear fashion, and

the explanatory variable regressions are linear as well. Therefore, nonlinear interactions

between the components of the model are ignored (Hasson and Heernan, 2011).

A DFA model has the following structure (Holmes et al., 2014):

yt = Zxt + a + vt where vt ∼ M V N (0, R) (3.9)

x0 ∼ M V N (π, Λ)

The general idea presented in Equation 3.9 is that the observed variables (y ) are mod-

eled as a linear combination of hidden trends (x). Then, the data entered into the model

(y ) is explained by some common trends (x). The factor loadings (Z ) are used, as in PCA,

to determine the variables that will be aggregated in each common trend. Other terms

in Equation 3.9 are matrices with the following denitions (see Holmes et al. (2014) for a

detailed explanation):

w is a m × T matrix of the process errors. The process errors at time t are multivariate

normal (MVN) with mean 0 and covariance matrix Q.

v is a n × T column vector of the non-process errors. The observation errors at time t are

a are parameters and are n × 1 column vectors.

Q and R are parameters and are m × m and n × n variance-covariance matrices.

π is either a parameter or a xed prior. It is a m × 1 matrix.

Λ is either a parameter or a xed prior. It is a m × m variance-covariance matrix.

There are three ways of estimating factor loadings in DFA: (i) use Maximum Likelihood

Function (MLE) and the Kalman Filter (KF); (ii) use Principal Components Extraction;

(iii) combination of the two rst. According to Montgomery and Runger (2003), MLE is

one of the best methods of obtaining a point estimator of a parameter. The estimator will

The Kalman lter is an algorithm for calculating the expected means and covariances

of the observed values for a whole time series in the presence of observation and process

error. In its original form it works only for models that are linear (exponential increase or

56 3. Literature on Performance Integration and Tools

decrease or expected constant population size over time) with multivariate normal error;

the extended Kalman lter uses an approximation that works for nonlinear population

Finally, interpretation of DFA results may not be straightforward. The DFA model uses

hypothetical latent variables (the common trends) that are deemed to be responsible for

are. Adding explanatory variables to the model could help with interpretation, but this

increases complexity and does not always improve the model. In general, one must keep in

mind that when using advanced techniques such as DFA, extra care may be needed when

3.4 Conclusions

This chapter is divided in two: the presentation of the literature about indicator rela-

indicators.

From the literature, the main conclusions we can take from papers are: the works carry-

ing out indicators aggregation does not use their results to achieve the global performance;

and, papers usually aggregate performance using tools which incorporate human judg-

These conclusions demonstrate a clear gap in the literature and this dissertation seeks to

aggregation.

To achieve our goal, it is necessary to investigate the statistical tools that are used for

dimension reduction. They are introduced in a summarized manner, since the objective of

the explanations is to allow the reader to recognize the characteristics and the requirements

to apply each technique. In Chapter 4, these methods are then evaluated according to the

requirements of the proposed methodology, determining the ones that can be used in our

studied problem.

The knowledge basis constructed in this chapter is used to develop our proposal method-

Chapter 4

Methodology to dene an Integrated Warehouse

Performance

experiments. Science is about explaining.

Bill Gaede

Contents

4.2 Conceptualization - The analytical model of performance

indicators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

4.3 Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

4.3.1 Data acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

4.3.2 Theoretical model of indicator relationships . . . . . . . . . . . . . . . . 63

4.3.3 Statistical tools application . . . . . . . . . . . . . . . . . . . . . . . . . . 64

4.4 Model Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

4.4.1 Integrated Performance proposition . . . . . . . . . . . . . . . . . . . . . 67

4.4.2 Scale denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

4.5 Implementation and Update . . . . . . . . . . . . . . . . . . . . . . . . 69

4.5.1 Integrated model implementation . . . . . . . . . . . . . . . . . . . . . . 69

4.5.2 Model update . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

4.6 Methodology implementation on this thesis . . . . . . . . . . . . 71

4.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

Abstract

This chapter presents the methodology to assess an integrated warehouse performance.

The methodology is divided in four main areas (conceptualization, modeling, model

solving, implementation and update), which are introduced in this chapter.

58 4. Methodology to dene an Integrated Warehouse Performance

According to Suwignjo et al. (2000), with the large number of multidimensional factors

aecting performance it is impossible to manage a scale system for each dierent dimension

of measurement. So, integrating those multidimensional eects into a single unit can

ology is introduced from a general point of view, being deeply detailed throughout the

sections.

research. For this kind of research, Mitro et al. (1974) propose the work development

the same four phases to present our developed methodology (Figure 4.1).

It is apparent in Figure 4.1 that the proposed methodology is dynamic. The imple-

mentation and update phase can be seen, at a rst glance, as the end of the methodology

application. However, if a situation changes in the warehouse, the proposed model needs

to be reviewed, and the methodology starts again by the conceptualization phase, closing

the loop.

performance indicators indicator relationships

• Model for indicators

aggregation

Conceptualization Modeling

Implementation

Model Solving

& • Determination of an

• Integrated model

implementation Update Integrated performance

• Model update model

• Scale definition

Figure 4.1: The proposed methodology phases with their main steps. Source: Adapted

from Mitro et al. (1974).

Figure 4.1 demonstrates the main outcomes inside of each phase in order to achieve

and implement the integrated performance model. The Conceptualization phase results

in an analytical model of performance indicators for the warehouse. Once the analytical

model is dened, the Modeling phase denes the relationships among indicators and how

they can be aggregated using dierent mathematical tools. Then, the Model Solving phase

analyzes the results obtained in the previous phase, proposing an integrated performance

model with a scale to evaluate and interpret the results. The last phase, Implementation

and Update, describes the integrated model implementation in a company as well as how

4.1. Introduction - General methodology presentation 59

to update it.

To perform the proposed methodology, Figure 4.2 shows the process ow, detailing the

steps carried out in each methodology phase (the dotted rectangles of Figure 4.2). Each

methodology application boundaries, i.e. in which warehouse areas the performance will

be measured and the indicators used for that. It means that, to perform the methodology,

it is necessary to dene the areas where the performance will be assessed and the indicator

set used by the company to achieve it. These indicators need to be known in terms of their

equations, since the analytical model is formed basically by this group of equations.

Conceptualization

Determination of indicator and data equations

Modeling

Theoretical model of indicator Model for indicators

relationships aggregation

Scale definition

Once the analytical model is developed (last step of conceptualization in Figure 4.2),

it is necessary to acquire data from indicators. This data is the time series of indicator

results, which are measured periodically in the enterprise. From this step, two analyses

and the use of historical data to perform indicators' aggregation. The theoretical model

is dened from the Jacobian matrix measurement, which is detailed in Section 4.3.2 and

60 4. Methodology to dene an Integrated Warehouse Performance

the indicators' aggregation are achieved from dimension-reduction statistical tools (Section

4.3.3).

From the results of the mathematical tools application, a quantitative model of indicator

provides as outcome the global warehouse performance. Because the performance values

necessary to create a scale for them. Finally, the implementation step demonstrates the

model utilization for periodic warehouse management and the update denes when the

indicators

The conceptualization phase involves the denition of the performance measurement sys-

tem. For this methodology purposes, it is necessary to perform three steps: the scope of

measurement, the denition of a metric set and, the determination of indicator equations,

It is really dicult to determine an evaluation model for distinct objectives, since each

enterprise (and, consequently, its warehouse) has specicities linked to dierent processes/

activities. Moreover, the performance measurement has become a strategic tool for cor-

porations to observe their weaknesses and act in a way to minimize them; so, it needs to

2009). Regarding the warehouse objectives, they are usually dened to improve the whole

supply chain performance, and this make the choice of an evaluation model crucial in a

networked organization. In fact, Fabbe-Costes (2002) states that all actors should create

value for chain partners; however, sometimes this is dicult to achieve because the actors

use dierent performance evaluation systems that are almost impossible to reconcile.

Besides the dierent warehouse objectives and processes, the performance measurement

systems should also satisfy some conditions such as (Manikas and Terry, 2010): inclusive-

ness (measurement of all related aspects), universality (allow for comparison under various

operating conditions), measurability (data required are measurable) and consistency (mea-

There are methodologies in the literature enabling to dene a set of performance indi-

cators based on strategic goals (Fernandes, 2006). Since the literature on this subject is

vast and the amount of indicators utilized in the process shall be carefully determined, the

denition of the indicators forming the warehouse metric system is out of this dissertation's

scope. In order to keep a large spectrum of applications for our methodology, we consider

that indicators utilized for warehouse management are derived from enterprise's strategy,

Regarding the steps of the conceptualization phase, we describe the approach as follows:

the performance will be measured. Kiefer and Novack (1999) state that the complexity

of the measurement systems increases as the number of activities performed by the ware-

4.2. Conceptualization - The analytical model of performance indicators 61

house increases. The proposed methodology considers that all warehouse activities can be

included in the measurement scope. However, the manager could have no interest in the

to determine the indicators set used for performance measurement. According to Melnyk

et al. (2004), the term metric is often used to refer to one of three dierent constructs:

(i) the individual metric; (ii) the metric set; and (iii) the overall performance measurement

systems. For the methodology application, the metric set is the group of indicators already

Steps 3 & 4: Even if some indicators from higher levels are generally related to the

ones of lower levels (Böhm et al., 2007), in this thesis we aggregate only the operational

metric set (i.e. the set of individual operational indicators). As our objective is to nd a

same does not happen to tactical and strategical indicators, which are usually related to

but some constraints must be satised: the indicators need to be measured in a quanti-

tative way, i.e. there are equations to describe them; historical data of measurement is

necessary to consider the indicator in the methodology, since this data will be used to

which are dened quantitatively by the equations:

house.

These quantitative indicators described in form of equations represent one part of the

analytical model. The second part comes from data equations. It is necessary to dene

data equations because sometimes collected data are calculated from other subdata, and

this information will be necessary to nd theoretically the relationship between indicators.

In our example, we consider that J data is calculated according to Equation 4.2, and all

J =A+G (4.2)

Thus, the nal analytical model for this example comprehends Equations 4.1 and 4.2.

The next section presents the modeling phase, which includes data acquisition, the

62 4. Methodology to dene an Integrated Warehouse Performance

4.3 Modeling

4.3.1 Data acquisition

The required data to apply the proposed methodology are time series of indicators, i.e.

indicator values measured periodically by the warehouse. We dene time series data as a

(du Toit and Browne, 2007). Initially, the number of measurements collected for the

same indicator (i.e. the dataset size) should be as long as possible and available in the

company. For instance, for the example described in Section 4.2, indicator's time series are

measured monthly in the warehouse as shown in Table 4.1, and the unit of each indicator

is demonstrated in parenthesis.

1 10 98 10 5 2 100

2 12 97 9 6 1.5 120

3 14 99 9 8 1 130

4 12 97 11 6 2 110

5 15 98 12 7 1.5 100

6 13 99 10 8 2 120

.. .. .. .. .. .. ..

. . . . . . .

Since the indicators are very heterogeneous with regard to their measurement units

($, time, %, etc.), Rodriguez et al. (2009) suggest three operations to be applied on raw

data: ltering, homogenization and standardization. The ltering analyzes the abnormal

behavior of the dataset; homogenization puts all data in the same temporal frequency (it is

necessary when some indicators are measured in weeks and others in months, for instance);

Xactual − Xmean

Xnew = (4.3)

σX

where Xnew is the new value of the variable, Xactual is the real variable value, Xmean is

the time series mean of the variable dataset, σX is the standard deviation of the variable

time series.

The nal dataset form a matrix of data ltered, homogenized in frequency and stan-

dardized, and ready for application of the proper mathematical techniques for identifying

relationships between indicators. The matrix is similar to Table 4.1, with the measurement

Depending on the statistical tool utilized (Section 4.3.3), there may be some limitations

For instance, some statistical tools may require the dataset to follow a normal distribu-

tion. To verify it, Newsom (2015) suggests to examine the skew and kurtosis of univariate

distributions. Kurtosis is usually a greater concern than skewness, but the literature only

recommends special analysis if skewness >2 and kurtosis > 7. If the univariate distribu-

tions are non-normal, the multivariate distribution will also be non-normal. One reason

4.3. Modeling 63

for non-normality is the presence of outliers in the dataset. In this case, the reason of the

outliers shall be examined, to eliminate the ones generated by typeset errors, for instance.

a rst moment it is not recommended to ll in the missing values by other ones generated

(for instance, there is a technique where the missing value is replaced by the time series

mean). As there are softwares to perform the statistical methods, usually they take care

automatically of this kind of issue, deleting the matrix line to eliminate the missing values.

Once the data is collected and treated, they are ready to be the inputs of mathematical

The quantitative relationships among indicators are the results from dierent variations

and eects of warehouse processes occurring at the same time. We could verify two main

forms of relationships: the eects of chained processes and of data shared among indicators.

The eect of chained processes is the impact of one performance indicator on the

other one that corresponds to the next activity in the process chain. For example, if an

order is shipped with delay, probably the delivery indicators (like delivery on time) will be

inuenced by this problem. So, one intervention in the system can cause a delay chain for

the rest of the process. However, the delay can be compensated by a great productivity of

the next operations, and at the end the order is delivered on time. Due to the variability of

the cases, this kind of relationship is not considered in the theoretical model construction.

The eect of data shared among indicators considers that two indicators are related

through the number of data they have in common. The main idea of this eect is that

if two indicators have in common one or more data, they have some kind of relationship

because once the data change, both indicators will be impacted, changing in some way.

This circumstance denes a relationship between two indicators. For example, labor pro-

ductivity and scrap rate use the same data, products processed, in their measurement

(Section 5.2 shows the indicator equations). If products processed change, both indicators

will also change. It is important to note that the variation intensity is not necessarily the

same in the concerned indicators. So, the data shared by indicators just suggest indicator

By the use of an analytical model, the data (and subdata) used in all indicator equations

can be easily veried. Thus, the analytical model dened in Section 4.2, with indicator

and data equations, is used as an input to assess indicator relations based on data sharing.

To certify the indicators which share similar data we calculate the Jacobian Matrix.

The Jacobian is a matrix of partial derivatives that is used to determine the out-

put/input relationship (Montgomery and Runger, 2003). In other words, the Jacobian is

a partial derivative matrix of the n outputs with respect to the m inputs. Each matrix

cell gives the sensitivity of the output with respect to one input variation, maintaining the

2007). To meet the methodology's purpose, we derive all functions f (indicator equations)

with respect to their data inputs x as shown in Equation 4.4.

64 4. Methodology to dene an Integrated Warehouse Performance

inputs

z

}| {

∂f1 ∂f1 ∂f1

...

outputs

∂x1 ∂x2 ∂xm

∂f ∂f2 ∂f2 ∂f2

∂x1 ∂x2 ... ∂xm

J= = . . .

(4.4)

∂x ..

. . . .

. . .

∂fn ∂fn ∂fn

∂x1 ∂x2 ... ∂xm

Equation 4.4 results in a n×m matrix where n is the outputs (indicators) and m the

number of inputs (independent data). The independent data refers to the non-combined

data used to calculate indicators. In the example dened in Section 4.2, the independent

data inputs used to assess the indicators (outputs) are A, B, C, D, E, F, G, H . The data

J is out of this list because it is calculated from the sum of A and G (Equation 4.2),

resulting in an aggregate data. For this example, the nal Jacobian Matrix is (after partial

A B C D E F G H

I1

I1

I2

I2

I3

I3

I4

I4

I5

I5

I6

I6

(4.5)

In the Jacobian matrix detailed in Equation 4.5, the non-zero cells signies that a

change in the data (input) will impact the indicator(s) (output). Therefore, it is possible

to identify the indicators which share data analyzing each matrix column. For example, the

column A has three non-zero cells, a11 , a41 and a61 , representing that this data inuence the

indicators 1, 4 and 6, respectively. Since these three indicators share data A, we conclude

Analyzing all data columns of the Jacobian matrix provide insights about indicator rela-

tionships in an innovative way, without considering human judgments nor possible dataset

issues when used in statistical analysis of relationships (since the dataset can contain im-

From the methodology steps of Figure 4.2, the development of the theoretical analysis

of indicator relations occurs in parallel with the model for indicators' aggregation. This

The objective of applying statistical tools is to group indicators based on their correlation

and data variation. The statistical tools available to achieve the objective of dimension-

tors is calculated from standardized time series data. The objective is twofold: correlation

4.3. Modeling 65

matrix is used as input of dimension-reduction tools and it gives a rst impression on the

It is important to emphasize that the relationship between the variables are described

as causal, meaning that it is explicitly recognized that a change of value in one variable

will lead to a change in another variable (Bertrand and Fransoo, 2002). However, it is not

possible to identify crossed relationships from correlation results, as this technique carries

out pair-wise comparisons between pairs of indicators instead of analyzing all indicators at

After obtaining the correlation matrix, statistical tools are applied to reduce data di-

tool has some requirements to allow its utilization. To assign data characteristics with the

mathematical tools requirements, Table 4.2 is built. It is divided in two parts: the right

side lists the requirements demanded by each method to be applied, according to the data

and sample characteristics presented on the left-side table. The objective of Table 4.2 is to

evaluate the suitable tools to be applied in the methodology, as shown on the last column

Data Characteristics

1. Data is a time series Could be used

Requisites to be

2. There are no missing values Math Tool in the proposed

applied

3. Data is non-stationary methodology?

4. Normality of data FA Item 2, 4, 5, 7, 8, 9 NO

5. Big sample size ( > 150) SEM Item 4, 5, 7, 9 NO

6. Small sample size CCA Item 4, 5 NO

7. Data is categorical PCA Item 2, 8 YES

8. Standardized data DFA Item 1, 3, 6 YES

9. Independence of observations

The mathematical tools analyzed in Table 4.2 are Factor Analysis (FA), Structural

FA and SEM demand a lot of data requirements, but the main issues impeding their

utilization are the big sample size (Item 5) and the independence of observations (Item

9). Regarding the sample size, Rodriguez-Rodriguez et al. (2010) arm that indicators

are dynamic as well as the PMS (Performance Measurement System), and enterprises

usually do not have large stores of data. They could keep nancial registers from lots of

years but it is not a normal practice for other PMS measures. The SEM method has a

measurement model less restrictive regarding the sample size. Fugate et al. (2010) state

that PLS (Partial Least Squares Regression - SEM measurement model) is often applied

for analyzing constructs because it accepts small sample sizes with no data distribution

requirements, as normality. However, our methodology proposes the use of time series as

model inputs, and this kind of data cannot t the condition of observations' independence.

also necessary to specify an initial model, i.e., to establish which are the observed variables,

66 4. Methodology to dene an Integrated Warehouse Performance

the error terms as well as their possible relationships (Rodriguez et al., 2009).

Therefore, FA and SEM are not initially considered as options for our methodology.

Even if there are some mathematical adjustments in FA and SEM model to overcome

the independence of observations problem, allowing their utilization with time series data

(for instance, see Choo (2004); du Toit and Browne (2007); Wang and Fan (2011)), these

In the case of CCA, the variables are initially classied in a specic group, and then,

the correlation between variables and groups are calculated in order to obtain the highly

arm that the CCA results are quite similar of PCA and DFA (i.e. to group variables

according to their similarities), we do not include the CCA as an option for our methodology

since there is no idea about which variables (in our case the indicators) can be classied on

the dependent and independent groups. Also, as big samples are required to apply CCA,

its utilization in our methodology are limited for the same reasons presented for FA and

SEM.

The techniques which can be used in our methodology are PCA and DFA. Both of

them could be used to determine indicator groups even if the techniques display some

dierences. PCA is optimal to nd linear combinations that represent the original set of

variables as well as possible, capturing the maximum amount of variance from the original

variables (Westfall, 2007). In the case of DFA, besides it is particularly designed for small

and non-stationary time series, this technique models the time series (variables) in terms

of a trend, seasonal eects, a cycle, explanatory variables and noise (Zuur et al., 2003a),

and the variables with similarities in these aspects are grouped together (called common

trend).

Getting back to the generic example started in Section 4.2, a possible result from the

dimension-reduction tool application (PCA or DFA) is illustrated in Figure 4.3. Taking the

indicators dened in Section 4.2, the dimension-reduction tool will separate them according

to their correlations.

Figure 4.3 shows a hypothetical result whereas the indicators are aggregated in two

dierent trends or components (the name will be in accordance to the tool applied). The

coecient of each indicator (α, β , γ , θ , η , µ) represents the relative weight between the

original variable (I1 , . . . , I6 ) and the component/trend. The results from Figure 4.3 can be

(4.6)

Component/T rend2 = θI3 − ηI5 + µI2

4.4. Model Solution 67

4.4.1 Integrated Performance proposition

This step of the proposed methodology comprehends the analysis of all mathematical

results achieving an integrated performance model at the end. The mathematical tools

analyzed are: Jacobian matrix , correlation matrix and PCA or DFA result.

Initially, the main objective is to dene which indicators should make part of the ag-

gregated model and which ones should be discarded. The theoretical model of indicator

relationships (the Jacobian matrix) and the correlation matrix may be evaluated together

to verify indicators that should be excluded because they will not t well the dimension-

reduction statistical tool. For instance, if the Jacobian matrix demonstrates that an indi-

cator does not share data with any other and, in the correlation matrix, the correlation

coecients r (named Person's r) are low (e.g. values lower than 0,3), we can conclude

that the indicator should be excluded from the model. After the exclusion of one indicator,

it is suggested to perform the PCA or DFA once again for the new indicator group.

In our theoretical example, the mathematical tools analyzed are the Jacobian matrix

(Equation 4.5) and the PCA/DFA result (Equation 4.6). From the Jacobian matrix we

conclude that indicator I3 has no relation with the indicator group, since it does not share

any data with other indicators. Moreover, let us consider that the correlation matrix

Therefore, both results (Jacobian and Correlation) recommend to discard this indicator

because it does not make part of the indicator's group which relates among them.

The exclusion of I3 requests a new application of PCA method (Figure 4.3 shows the

rst result), which hypothetically has the following new outcome (Figure 4.4 and Equation

4.7). The exclusion of I3 has improved the result, since all the indicators are explained now

by just one component/trend in the new outcome. Equation 4.7 shows the nal result. It is

important to note that the indicators' coecients have also changed due to the modication

Equation 4.7 represents the integrated performance model for the example carried out

throughout this chapter. However, if the number of indicators is high, usually it is dicult

to aggregate all measures in just one component. Generalizing the result of Equation 4.7,

68 4. Methodology to dene an Integrated Warehouse Performance

a generic model representing several components can be described by Equation 4.8, from

Manly (2004).

m

X

Ci = bij Xj ∀ i = 1, . . . , n (4.8)

j=1

where:

Ci = principal components

b11 . . . bnm ∈ < = relative weight of each variable (X1 , . . . , Xm ) in the corresponding

component (C1 , . . . , Cn )

X1 . . . Xm = performance indicators.

the company and used for daily management. For that, each component needs a scale to

allow the interpretation of results, since the inputs are normalized indicators, producing

components without units. Analyzing the global performance of a warehouse using dif-

ferent components without physical units can be dicult for managers because of their

subjectivity.

are several methods to achieve this global expression. Lohman et al. (2004) state that

aggregation can be done directly if the underlying metrics are expressed in the same units of

measure, which can be achieved after a normalization, for example. Clivillé et al. (2007) cite

some examples of methods as the weighted mean, which is the more common aggregation

operator; the weighted arithmetic mean; and the Choquet integral aggregation operator,

which generalizes the weighted mean by taking mutual interactions between criteria into

account.

Taking the components of Equation 4.8 and aggregating them using the weighted mean,

GP = a1 × C1 + a2 × C2 + a3 × C3 + . . . + an × Cn (4.9)

where:

C1 . . . Cn = principal components which group X1 up to Xm in linear combinations.

The determination of the component weights depends on several factors. Firstly, de-

pends upon the aggregation formula. For example, the criteria in a weighted mean and

in a weighted geometric mean would not be the same (Clivillé et al., 2007). Secondly, the

relative importance of the indicators should be considered. Each warehouse will have dif-

ferent results of indicator aggregation which requires an analysis of the indicators grouped

in each component to dene their weights (some indicators could be more important than

others). Lastly, the weights depend upon the warehouse strategy. Each warehouse has

dierent objectives and can rank component's weight according to its priority.

The company can choose to stop the solution method in the component level (Equation

4.8) or build the integrated indicator (Equation 4.9). In both situations, the results of

4.5. Implementation and Update 69

A scale determines the maximum and minimum values reached by a variable. It can be

used to develop interview instruments in an organized way, verifying some hypothesis from

the data. For example, Chen (2008) uses a six-item scale to measure the operational

However, the scale is used in our work as a reference point to evaluate the results of given

variables, which are the principal components (Equation 4.8) and the integrated indicator

(Equation 4.9).

Jung (2013) states that the four main types of scale are: nominal scales (categorical:

only attributes are named), ordinal scales (rankings: attributes can be ordered), interval

scales (equal distances corresponding to equal quantities of the attribute), and ratio scales

(equal distances corresponding to equal quantities of the attribute where the value of zero

corresponds to none of the attributes). The scale developed for our purpose is the interval

Regarding the dierent measurement units of indicators (time, %, etc.), Rodriguez et al.

(2009) propose the auto-scaled technique, which combines centering and standardization.

The scale is built for each variable independently, using its mean and standard deviation

to dene the lower and an upper scale limits. One potential problem of the auto-scaled

technique is that it does not allow the comparison among dierent variables, because each

The work of Lohman et al. (2004) proposes the normalization method to create the same

scale range for dierent indicators. The authors determine a linear 0 − 10 scale. Two steps

need to be taken for normalizing the metric scores (Lohman et al., 2004): (1) the denition

of the metric score range that corresponds to the 0 − 10 scale; (2) the normalization of

the scores to a 0 − 10 scale, since the values 0 − 10 should always have the same meaning,

For the component expressions (Equation 4.8) it is not possible to use this kind of

procedure since indicators may have opposite objectives (e.g. the productivity wants high

values whereas time aims for the lower ones), complicating the target denition.

One possible solution is the use of optimization methods to dene the best warehouse

performance. It facilitates the inclusion of dierent indicator goals in the same model as

The proposed scale using optimization seems a good option to evaluate the integrated

Chapter 7.

4.5.1 Integrated model implementation

The implementation consists of demonstrating the equations that may be maintained and

refreshed for periodic management, and how the integrated results should be interpreted.

70 4. Methodology to dene an Integrated Warehouse Performance

The expressions that will be used by the manager for warehouse performance measure-

ment are Equations 4.8 and 4.9. Other equations from the analytical model are not used

once the aggregated model is achieved. It is important to note that the coecients `ai ',

The objective of the integrated model is to be used as any other indicator system,

being measured periodically and analyzed according to a given objective. To attain this,

component values;

4. These component results are used in Equation 4.9 to obtain the integrated indicator

value.

This procedure should be done periodically (preferably with the same periodicity of

the performance indicator measurements) allowing to follow the evolution of the integrated

indicator throughout time. As all operational performance indicators are also measured,

it is possible to identify signicant changes in indicators which alter the aggregated one.

Moreover, the developed scale provides the warehouse performance limits; if the manager

progress.

Before the implementation, it is important to conrm with the managers that the results

from the aggregated model and the scale t the warehouse reality. If it is conrmed, the

analytical model and aggregated performance expressions are validated by the reality.

The aggregated model cannot be considered as a static entity: it must be maintained and

updated to remain relevant and useful for the organization (Lohman et al., 2004). However,

some authors cite that the literature has not yet satisfactorily addressed the issue of how

performance measures should evolve over time (i.e. be exible) in order to remain relevant

with the constant evolution of organizations (Kennerley and Neely, 2002; Neely, 2005).

Regarding this situation, our methodology proposes a periodic reevaluation of the in-

tegrated performance model. This reevaluation encompasses mainly the selection of the

metrics with their equations, the application of statistical tools (PCA, DFA) in a new

dataset to compare the results, and the revision of component's weights in the integrated

indicator equation.

The aggregated performance model, which emerges from the proposed methodology,

has a life cycle and is only valid as long as the internal and external environment remains

stable. For example, new business areas or new challenges require a revision of the model. A

periodic revision of the model can help with this identication. It is important to recognize

these changes as soon as possible to redene the quantitative basis of the model. This

4.6. Methodology implementation on this thesis 71

practice is also used in other PMS proposed in the literature (e.g. Suwignjo et al. (2000)).

The model redenition encompasses the comparison of desired performance indicators with

existing measures (to identify which current measures are kept, which existing measures

are no longer relevant, and which gaps exist so that new measures are needed) (Lohman

et al., 2004).

The methodology presented in this chapter explains the steps that should be done to at-

cases, a specic one to be applied. That is the case of: indicator equations denition (each

warehouse should dene its indicator set); the dimension-reduction statistical tool (PCA

or DFA); the nal aggregated model (using just component equations or also the aggre-

gated indicator), the criterion for scale optimization (the constraints depend on warehouse

situation). Therefore, the methodology provides a customized integrated model for each

overview of the methodology application on this thesis. Figure 4.5 depicts the activities

performed (numbered from 1 up to 21) to achieve the aggregated model for the studied

warehouse.

Figure 4.5 shows the phases of the methodology in green dotted rectangles (Concep-

tualization, Modeling, Model Solving, Implementation and Update). The blue rectangles

highlight the main outcomes of each phase with their activities written in black.

that we dene a standard warehouse (activity 1) as the object of the study, and the

performance indicators are taken from the literature to manage the ctitious warehouse.

The equations for these indicators are a result of the literature interpretation, based on

metric's denition (activity 2). The nal group of equations form the analytical model,

r (Component Architecture for the Design of

Engineering Systems). We use this software to analyze the analytical model, providing the

independent inputs and outputs (activity 3) and calculating the Jacobian matrix (activity

6).

indicators periodically reproducing the warehouse dynamics (activity 4). This data is used

for the next steps: theoretical model of indicator relations and aggregated model. For

the rst one, just a sample is used to calculate the Jacobian matrix automatically using

CADES

r (activity 5). For the second one, the dataset created is standardized (activity 9)

Regarding the application of statistical tools, the correlation matrix is calculated (activ-

ity 8) as well as the PCA method (activity 10). Both results with the indicator relationships

matrix (activity 7) are analyzed to provide insights about the behavior of the indicators.

The PCA and DFA have been tested to aggregate indicators. For the DFA, results do not

t with the objective of this thesis. Further study needs to be developed and it is proposed

as future research (for details about the rst results obtained see Appendix F). In the case

72 4. Methodology to dene an Integrated Warehouse Performance

2. Determination of indicator and data equations

Analytical

model

based on the literature Conceptualization

3. The analytical model is coupled with CADES

software. Result: inputs and outputs identification

Acquisition indicators

Modeling

5. Determination of Inputs values

Theoretical 8. Assessment of indicators’ correlation

6. Jacobian matrix calculation using CADES Model for

model of matrix

software. The result is a matrix of relations indicators’

indicator between data (inputs) and indicators (outputs) 9. Indicator standardization

relationships aggregation

7. Matrix of indicators’ relationships based 10. Statistical tools application: PCA

on the number of data in common

Analysis of Partial Results

Theoretical(7) x Correlation(8) x 12. Factors retained performance

Statistics (10) 13. Set of indicators related

proposition Model Solving

15. Factors weights 16. Objective function 18. Inputs/outputs limits Scale definition

17. Adjustment of the 19. Optimization

14. Constraints 20. Integrated indicator scale

analytical model (3) for optimization. tool choice

21. We explain how to implement and interpret the results obtained as Implementation and

well as its update when necessary. Update

of PCA, good results are attained and it was the chosen method to determine indicator

groups. Moreover, it is simple to apply and interpret, which are interesting characteristics

The partial results (7, 8 and 10) are analyzed to dene the integrated performance

model with a global indicator (activities 11, 12, 13). The integrated indicator scale is

dened using an optimization model (activities 14 up to 20). The application nishes with

an explanation of the model utilization for periodic management and when its update is

4.7 Conclusions

This chapter presents a methodology to dene an integrated warehouse performance model.

It consists of several steps to analyze indicator relationships from dierent points of view,

using distinct mathematical tools to group these indicators according to their correlation

model: the analytical model and the Jacobian matrix measurement to analyze indicator

4.7. Conclusions 73

relationships; the statistical tools to propose indicator groups; the optimization method to

develop the scale for the integrated indicator. This multidisciplinary approach permits a

The methodology is general; it gives several alternatives that one can choose when

developing the integrated model. Each warehouse can present dierent objectives, pro-

cesses, particularities, and the fact of not specifying the tools allows the adaptation of the

Chapter 5

Conceptualization

The question is not what you look at, but what you see.

Henry David Thoreau

Contents

5.1.1 Warehouse Layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

5.1.2 Measurement Units of Performance Indicators . . . . . . . . . . . . . . . 77

5.2 Analytical model of Indicator Equations . . . . . . . . . . . . . . 77

5.2.1 Denition of the metric set . . . . . . . . . . . . . . . . . . . . . . . . . . 77

5.2.2 Transformation of Indicator Denitions in Equations . . . . . . . . . . . 80

5.2.3 Notation to describe Indicator Equations . . . . . . . . . . . . . . . . . . 81

5.2.4 Time Indicators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

5.2.5 Productivity Indicators . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

5.2.6 Cost Indicators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

5.2.7 Quality Indicators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

5.3 Complete Analytical Model of Performance Indicators and

Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

5.3.1 The Construction of Data Equations . . . . . . . . . . . . . . . . . . . . 92

5.3.2 Analytical model assumptions . . . . . . . . . . . . . . . . . . . . . . . . 94

5.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

Abstract

This chapter performs the Conceptualization phase of the methodology. It begins by

presenting the studied standard warehouse, with its characteristics and processes (i.e.,

the scope of the work). Thereafter, the metric system used to measure the warehouse

performance is dened, based on the literature review. To determine the rst part of the

analytical model, formed by indicator equations, indicator denitions are interpreted

in detail. Finally, the data of all indicators are expanded into new equations; then, the

complete analytical model is constituted of indicators and data equations.

76 5. Conceptualization

This chapter is the starting point to implement the methodology presented in Chapter 4,

establishing the basis of an integrated performance model. All steps needed to develop

an analytical model of performance indicators are carried out: the denition of the per-

formance measurement scope; the determination of the indicator set; the formulation of

customer requirements, service level oered, etc. The scope of this implementation is on a

hypothetical warehouse, named standard warehouse (shown in Figure 5.1). The denomina-

tion standard is due to the processes carried out on it. We consider the main operational

activities performed by the majority of warehouses, which are (Section 2.3.4 presents their

denitions): receiving, storing, internal replenishment, order picking, shipping and deliv-

ery. Thus, the performance measurement is carried out on the warehouse shop oor, also

Figure 5.1 details not only the boundaries of the activities carried out in the stan-

dard warehouse but also its layout and the measurement unit limits of the performance

The layout of the standard warehouse is shown in the middle part of Figure 5.1 with the

following regions: receiving docks for truck assignment, unloading area, inventory area,

Since the majority of warehouses have intensive handling activities in order picking

(De Koster et al., 2007), this warehouse follows a manual system for storing and picking

products. In the manual system, the order picker/forklift driver has to store products in a

5.2. Analytical model of Indicator Equations 77

proper location (in case of storage activity) or localize and pick the searched products in

We consider that this facility supplies the market with a make-to-stock production. In

a make-to-stock operation, the customer orders launch a process in the picking area, going

The inventory area of Figure 5.1 comprehends the reserve storage area and the forward

picking area. The reserve area contains the bulk stock and it is located in superior rack

levels. The forward picking area is situated in the same racks of bulk stock, but in the

inferior levels to facilitate order picking process. So, this conguration implies in regular

The inbound area of the warehouse encompasses the receiving of trucks until the storage

of products in inventory area, and the outbound area comprises the replenishment activity

performed from the inventory area up to the delivery of the product to the client.

The top of Figure 5.1 demonstrates the boundaries of measurement units used to calculate

warehouse performance indicators in this dissertation. The units are: pallets, order lines

and order.

A customer order or simply order (as described in this work) is an individual cus-

and the quantity of each one (Johnson et al., 2010). Order lines are the number of dif-

ferent product types in a customer order. Each line designates a unique product or stock

keeping unit (SKU) in a certain quantity (De Koster et al., 2007). A pallet refers to the

products transported on it, with the quantity and kind of products varying from one pallet

to another.

Each measurement unit described in the top of Figure 5.1 is related to the indicator

units in one or more warehouse activities. For example, in receiving, storage and internal

replenishment, the operations are measured in pallets. Similarly, order lines is the unit

for picking indicators and order is the standard measure for delivery indicators.

The exception is the shipping activity, where both order lines and order are used

to measure shipping indicators. Packing and shipping are transition areas, in which some

indicators are related to internal operations (e.g. labor performance in shipping activity)

As each part of the warehouse uses a specic unit of measure (for instance, pallets, or-

ders), we also dene a smaller unit related to a single item, named product or SKU(stock

keeping unit). This distinct notation is used in more general indicators, measuring sev-

eral activities (e.g., Stock out rate, Equation 5.40) or the whole warehouse (e.g., Labor

5.2.1 Denition of the metric set

After the denition of the warehouse characteristics, the metric system used for perfor-

mance measurement needs to be dened. Keebler and Plank (2009) study the logistics

78 5. Conceptualization

measures most commonly used by the managers in the US industry. The results show some

preference for the indicators such as the outbound freight cost, the inventory count accu-

racy, the nished goods inventory turn and the order ll. However, the authors conclude

that there is not a consensus of a group of measures used to assess warehouse performance.

The methodology presented in Chapter 4 determines that the indicators used to develop

the integrated model need to come from strategic goals of the enterprise. As our standard

warehouse is theoretical, we consider that its operational metric system comes from the

Regarding the indicator requisites, the methodology denes that they need to be quan-

titatively measured, i.e. it is necessary to describe them in equations. Thus, the metric

system is dened from the direct indicators resulting from the literature review, which are

Comparing Tables 2.12 and 2.13 with the warehouse characteristics, we can see that

not all warehouse areas contain indicators (e.g. there are no indicators related to the

replenishment). Moreover, some indicators related to specic activities are missing. That

is the case of productivity indicators, for example. Table 2.12 shows productivity indicators

Therefore, we make some adjustments in the initial group of indicators taken from the

literature which result in the indicators presented in Table 5.1. To maintain consistency

among the warehouse activities, indicators related to internal replenishment, not veried

in literature review but also important for warehouse management, are added to the metric

set. Furthermore, quality and productivity indicators for receiving and storage activities

From the literature, we can infer that the cost indicators are not so frequently used for

in papers are more global and usually related to several activities, demonstrating that

costs are analyzed in managerial levels. One reason for that could be that the operational

objectives of the warehouse are usually related to process performance due to the intensive

work-handling (e.g. lead time reduction, quality improvements) instead of cost measures.

For these reasons, we have not included new cost indicators related to specic activities as

5.2. Analytical model of Indicator Equations

Table 5.1: Final warehouse performance indicators group.

Receiving Storing Inventory Replenishment Picking Shipping Delivery

Time Rect Putt Rept * Pickt Shipt Delt

Invq , Shipq , Delq ,

Quality Recq * Stoq Repq * Pickq

StockOutq OTShipq OTDelq

Cost Invc Trc

InvUtp , Delp *,

Productivity Recp Stop * Repp * Pickp Shipp

TOp TrUtp

Dimension Process - Transversal Indicators

Inbound Process Outbound Process

DSt OrdLTt

Time

OrdFq , PerfOrdq

Quality

CustSatq , Scrapq

OrdProcc

Cost

CSc , Labc , Maintc

Productivity

Thp , Labp , WarUtp , EqDp

∗ The symbol denotes indicators added after the literature review analysis.

79

80 5. Conceptualization

In contrary of the indicator additions, some others listed in the metric system of Chapter

2 are not included in Table 5.1 since more general metrics encompass them. That is the case

of Queuing time and Outbound space utilization. For Queuing time, it is comprised in

data equations of time indicators (Appendix A demonstrates this parameter inside the time

The nal warehouse metric system analyzed in this thesis has 41 indicators. Table

5.1 shows these indicators using the same table format presented in Chapter 2. The only

dierence is that, besides the metrics added (highlighted with the symbol *), the resource

related indicators (Labor cost `Labc ', Labor productivity `Labp ', Equipment downtime

`EqDp ', Maintenance cost `Maintc ' and Warehouse utilization `WarUtp ', presented in

Table 2.13) are also included in Table 5.1, being classied as transversal indicators.

The notation used in Table 5.1 to describe indicators is a standard created in this thesis

Finally, it is important to underline that this group of indicators does not provide an

can be measured by the warehouses which are not included in Table 5.1.

After the determination of the nal group of indicators, their denitions are used as a

basis to establish indicator equations. While some denitions are easily transformed in

equations, others do not have the same interpretation. The denitions come basically from

the same paper database of the literature review. In the cases that the indicators are not

dened in papers, we look for these denitions in a supplementary database. Tables 5.2,

5.4, 5.6, 5.8 present three kinds of indicators distinguished by the symbols

a , b and c . The

indicators symbolized as

a need an interpretation of their denitions in order to transform

them into equations. One example is receiving time indicator dened as unloading time

(see Table 5.2). We determine its equation as the total unloading time divided by the

number of pallets unloaded in a month (Equation 5.1). The indicators represented by the

symbol

b are the ones for which neither the denition nor the measurement are found in the

literature. We dene these indicators based on the best common sense that we could infer

c is attributed to maintenance cost indicator (Table 5.6),

the only metric dened by the union of two distinct denitions (from De Marco and Giulio

(2011) and Johnson et al. (2010)). In the cases where there is more than one denition,

they are demonstrated in the table (e.g. order lead time in Table 5.2) and the alternatives

All other indicators, described in Tables 5.2, 5.4, 5.6, 5.8 without symbols, have their

measurement given directly by their denition (e.g. lead time to pick an order line, total

of products stored per labor hour storing, etc.). Some of these denitions are just adjusted

to the measurement unit used in this work. For example, picking accuracy is dened as

orders picked correctly per orders picked but we changed the unit order picked to order

line picked.

5.2. Analytical model of Indicator Equations 81

The nal metric system encompasses Equation 5.1 to Equation 5.41. To better illustrate

the results, we show in parenthesis the equation outcomes, even if they are not units

derived from International System of Units (SI). For example, we dene pallet as a

pseudo unit indicating the number of pallets. To dene the data used in each indicator's

equation, Tables 5.3, 5.5, 5.7, 5.9 describe the data meanings and their measurement units

(in parenthesis). The time base used in this work is month and the measurement unit

The indicator notations presented in this chapter are used all along the thesis. All

indicator names are written in bold format (for instance, Rect ) and data used in indicator

equation are in sans serif style (e.g. Pal Unlo). Moreover, the indicators have also a letter at

the end of the indicator name to designate their classication: t for time, p for productivity,

c for cost and q for quality.

The next sections present the indicator equations separated in terms of time, produc-

The time indicator equations are elaborated from the interpretation of the indicator def-

initions given in Table 5.2. The data used in these time indicators (Equation 5.1 up to

Table 5.2 presents two indicators with more than one interpretation: order lead time

and dock to stock time. Analyzing order lead time denition from customer's perspective,

it should encompass from the time when the customer order is placed up to the time when

the customer receives his order and not until the product is shipped by the warehouse.

Thus, all parts of the supply chain involved to the accomplishment of this task should

be included in this indicator. For dock to stock time, it is important to note that some

denitions could be misleading. The denition of Ramaa et al. (2012) could be interpreted

as if the indicator comprehends the inventory and replenishment times (time from the

storage up to the product is picked), but this is not the case. The authors consider that

the product is available for order picking at the moment of storing. Therefore, dock to

stock is the time from supply arrival up to the storage in the inventory oor.

Usually, the activities performed in a warehouse are sequential, i.e. the shipping starts

after the picking is nished. As the time indicators are measured in terms of the mean time

one activity takes, it is possible to depict all these measures in a timeline, as shown in Figure

5.2. Some events are pointed out in the timeline, to demonstrate exactly the beginning

and the end of each measurement, according to the denitions described in Table 5.3. It is

important to highlight that ∆t(Rec) encompasses also the inspection activity, which takes

some time after the unloading nishes to enable the pallets to be stored.

The time indicators (Equation 5.1 up to Equation 5.8) are measured monthly, so the

sum operator in all equations are related to the activities performed during a whole month.

The indexes p, l and o in indicator equations correspond to pallets, order lines and orders,

respectively.

82 5. Conceptualization

Gu et al. (2007);

Receiving

Rect unloading time Matopoulos and (5.1)a

time

Bourlakis (2010)

lead time between the

Mentzer and

product(s) is unloaded

Konrad (1991);

Putaway and available to be

Putt De Koster et al. (5.2)a

time storage until its eec-

(2007); Yang and

tive storage in a desig-

Chen (2012)

nated place

lead time from sup-

Dock to

ply arrival until prod-

stock Ramaa et al. (2012)

DSt uct is available for or- (5.3)a

time

der picking

the amount of time it

takes to get shipments

Yang and Chen

from the dock to in-

(2012)

ventory oor without

inspection

lead time to transfer

Replenish-

products from reserve Manikas and Terry

Rept ment (5.4)

storage area to for- (2010)

time

ward pick area

Order

lead time to pick an Mentzer and Kon-

Pickt picking (5.5)

order line rad (1991)

time

lead time to load a

Shipping

Shipt truck per total orders Campos (2004) (5.6)

time

loaded

total time of distribu-

Delivery

Delt tions per total orders Campos (2004) (5.7)

lead time

distributed

Mentzer and Kon-

rad (1991); Kiefer

lead time from cus- and Novack (1999);

Order

tomer order to cus- Rimiene (2008);

OrdLTt lead time (5.8)a

tomer acceptance Menachof et al.

(2009); Yang and

Chen (2012)

lead time from or-

Yang (2000); Ra-

der placement to ship-

maa et al. (2012)

ment

a

Interpretation of the indicator denition

5.2. Analytical model of Indicator Equations 83

Time

Δt(DS) Δt(Ord)

P alU

Pnlo

∆t(Rec)p

p=1 hour

Rect = ( pallet ) (5.1)

Pal Unlo

P alSto

P

∆t(Sto)p

p=1 hour

Putt = ( pallet ) (5.2)

Pal Sto

P alSto

P

∆t(DS)p

p=1 hour

DSt = ( pallet ) (5.3)

Pal Unlo

P alM

Poved

∆t(Rep)p

p=1 hour

Rept = ( pallet ) (5.4)

Pal Moved

OrdLiP

P ick

∆t(Pick)l

l=1 hour

Pickt = ( order line ) (5.5)

OrdLi Pick

OrdLiShip

P

∆t(Ship)l

l=1 hour

Shipt = ( order line ) (5.6)

OrdLi Ship

OrdDel

P

∆t(Del)o

o=1 hour

Delt = ( order ) (5.7)

Ord Del

OrdDel

P

∆t(Ord)o

o=1 hour

OrdLTt = ( order ) (5.8)

Ord Del

84 5. Conceptualization

Notation Denition

Time between the truck assignment to a dock and the moment when the unloading

∆t(Rec) =

nishes and the pallet is available to be stored (hour/P alU nlo)

Time between the instant when the pallet is available to be stored and its eective

∆t(Sto) =

storing (hour/P alSto)

Time between the truck assignment to a dock up to the storing of the pallet

∆t(DS) =

(hour/P alU nlo)

Time to transfer a pallet from the reserve storage area to the forward picking area

∆t(Rep) =

(hour/P alM oved)

Time between the instants when operator starts to pick an order line and when the

∆t(Pick) =

picking nishes (hour/OrdLiP ick)

Time between the instants when the order picking nishes and when the order line

∆t(Ship) =

shipping is loaded in the truck (hour/OrdLiShip)

Time between the truck leaving the warehouse and the customer acceptance of the

∆t(Del) =

product (hour/OrdDel)

Time between the customer ordering and the customer acceptance of the product

∆t(Ord) =

(hour/OrdDel)

Pal Unlo = number of pallets unloaded per month (pallets/month)

Pal Sto = number of pallets stored per month (pallets/month)

Pal Moved = number of pallets moved during replenishment operation per month (pallets/month)

OrdLi Pick = number of order lines picked per month (order lines/month)

OrdLi Ship = number of order lines shipped per month (order lines/month)

Ord Del = number of orders delivered per month (orders/month)

Productivity can be dened as the level of asset utilization (Frazelle, 2001), or how well

resources are combined and used to accomplish specic, desirable results (Neely et al.,

work completed, and/or services produced and quantities of inputs or resources utilized to

One of the most commonly used productivity measure is the labor productivity. Indeed,

warehouses usually have many handling-intensive activities. Bowersox et al. (2002) arms

that logistics executives are very concerned with labor performance. In fact, the number

of papers found concerning this theme conrms his statement. There are several ways

to measure labor productivity, and two denitions are presented in Table 5.4. The rst

labor productivity indicator (from De Marco and Giulio (2011)) measures the workers'

eciency, verifying the production during the real time used to execute the tasks. The

second denition (from Frazelle (2001)) produces a measure based on the work done during

the available time to work, e.g. measuring the number of items processed during a day.

We use the last indicator in our work because it is the most commonly used in warehouses

denition in Table 5.4 and Equation 5.9 shows that this indicator does not measure directly

the employee eciency, it focuses on time usefulness. It means that all incoming ow

5.2. Analytical model of Indicator Equations 85

ratio of the total num-

ber of items man-

Labor pro- De Marco and Giulio

aged to the amount of

Labp ductivity (2011) (5.9)

item-handling working

hours

total produced per to-

Frazelle (2001)

tal man-hour

Receiving number of vehicles un- Mentzer and Konrad

Recp (5.10)

productivity loaded per labor hour (1991)

total number of prod-

Storage pro- ucts stored per labor

Stop our denition (5.11)b

ductivity hour in storage activ-

ity

total number of pallets

Replenishment moved per labor hour

Repp our denition (5.12)b

productivity in replenishment activ-

ity

total number of prod- Kiefer and Novack

Picking pro- ucts picked per labor (1999); Manikas and

Pickp (5.13)

ductivity hours in picking activ- Terry (2010); Yang and

ity Chen (2012)

Mentzer and Konrad

total number of prod-

Shipping pro- (1991); Kiefer and No-

Shipp ucts shipped per time (5.14)

ductivity vack (1999); De Koster

period

and Waremius (2005)

total number of or-

Delivery Pro- ders delivered per la-

Delp our denition (5.15)b

ductivity bor hours in delivery

activity

Inventory rate of space occupied Ramaa et al. (2012);

InvUtp (5.16)

utilization by storage Ilies et al. (2009)

ratio between the cost Johnson and McGinnis

TOp Turnover of goods sold and the (2011); Yang and Chen (5.17)

average inventory (2012)

O'Neill et al. (2008);

Transport

TrUtp vehicle ll rate Matopoulos and (5.18)

utilization

Bourlakis (2010)

Warehouse rate of warehouse ca-

WarUtp Bowersox et al. (2002) (5.19)

utilization pacity used

percentage of hours

Equipment

EqDp that the equipment is Bowersox et al. (2002) (5.20)

downtime

not utilized

Mentzer and Konrad

(1991); Gunasekaran

and Kobu (2007);

items / hour leaving Kiefer and Novack

Thp Throughput (5.21)

the warehouse (1999); De Koster and

Waremius (2005);

Voss et al. (2005); Gu

et al. (2007)

b

This indicator is not explicitly dened in the literature and we consider the denition

presented in this table for the purpose of this work.

86 5. Conceptualization

(number of products processed per month, in our case) processed will be divided by the

total number of hours available to work. If there are some periods where there is no product

to process, this will reduce the indicator result even if the employees have worked well.

The indicator Equipment Downtime, EqDp Equation 5.20, was initially identied in

the work of Mentzer and Konrad (1991) and dened as a period in which an equipment

is not functional, downtime incurred for repairs. Since this is a time indicator, they

were classied in this dimension in Section 2.4.1. However, the denition of Bowersox

et al. (2002) presented in Table 5.4 produces an indicator with more information, relating

the time in which the equipment is not functional in all available time. For this reason,

interesting to highlight that the pseudo unit times, in Equation 5.17, signies the number

Labp = ( hour ) (5.9)

WH

Recp = ( ) (5.10)

WH Rec hour

Stop = ( ) (5.11)

WH Sto hour

Repp = ( hour ) (5.12)

WH Rep

Pickp = ( hour ) (5.13)

WH Pick

Shipp = ( hour ) (5.14)

WH Ship

Delp = ( ) (5.15)

WH Del hour

Inv CapUsed

InvUtp = × 100(%) (5.16)

Inv Cap

CGoods

TOp = (times) (5.17)

Ave Inv

Kg Tr

TrUtp = × 100(%) (5.18)

Kg Avail

War CapUsed

WarUtp = × 100(%) (5.19)

War Cap

HEq Stop

EqDp = × 100(%) (5.20)

HEq Avail

5.2. Analytical model of Indicator Equations 87

Notation Denition

Ave Inv = average warehouse inventory per month ($/month)

CGoods = cost of all products sold by the warehouse per month ($/month)

Inv CapUsed = average number of pallets in inventory per month (pallets/month)

Inv Cap = total amount of pallet space (pallets)

Kg Tr = total of kilograms transported per month (kg/month)

Kg Avail = delivery capacity in kilograms per month (kg/month)

total number of hours during which equipments are stopped per month

HEq Stop =

(hour/month)

total number of hours during which equipments are available to work per month

HEq Avail =

(hour/month)

OrdLi Pick = number of order lines picked per month (order lines/month)

OrdLi Ship = number of order lines shipped per month (order lines/month)

Ord Del = number of orders delivered per month (orders/month)

Pal Unlo = number of pallets unloaded per month (pallets/month)

Pal Sto = number of pallets stored per month (pallets/month)

number of pallets moved during replenishment operation per month

Pal Moved =

(pallets/month)

Prod Ship = number of products shipped per month (nb/month)

number of products processed by the warehouse per month. Products processed

Prod Proc =

refers to the number of products shipped in the warehouse (products/month)

total item-handling working hours for all warehouse activities per month. In this

WH = thesis, WH is calculated by the sum of WH Rec, WH Sto, WH Rep, WH Pick, WH

Ship (hour/month)

War CapUsed = total warehouse oor area occupied by activities per month (m2 /month)

War Cap = total warehouse capacity oor (m2 )

total employee labor hours available for receiving activity per month

WH Rec =

(hour/month)

WH Sto = total employee labor hours available for storing activity per month (hour/month)

total employee labor hours available for replenishment activity per month

WH Rep =

(hour/month)

WH Pick = total employee labor hours available for picking activity per month (hour/month)

WH Ship = total employee labor hours available for shipping activity per month (hour/month)

WH Del = total employee labor hours available for delivery activity per month (hour/month)

total number of hours during which the warehouse is open per month

War WH =

(hour/month)

88 5. Conceptualization

the holding cost

Inventory

and the stock out Li et al. (2009)

costs

Invc penalty (5.22)a

total storage costs

Rimiene (2008)

/ unit

Cagliano et al.

inventory level

(2011); Gallmann

(measured mone-

and Belvedere

tarily)

(2011)

amount of dollars

Transportation Bowersox et al.

Trc spent per order de- (5.23)

costs (2002)

livered.

total processing

Order pro-

cost of all orders

OrdProcc cessing Campos (2004) (5.24)

per number of

cost

orders

total warehousing Bowersox et al.

Cost as a % cost as a percent (2002); Ilies et al.

CSc (5.25)

of sales of total company (2009); Ramaa

sales et al. (2012)

cost of personnel

Cagliano et al.

Labc Labor cost involved in ware- (5.26)

(2011)

house operations

costs of building (1)- De Marco

Maintenance maintenance (1) and Giulio (2011)

Maintc (5.27)c

cost and equipment (2)- Johnson et al.

maintenance (2) (2010)

a

Interpretation of the indicator denition or many indicators' aggregation c Union

of two distinct denitions.

Thp = ( hour ) (5.21)

War WH

In Table 5.6 there are three dierent denitions for inventory costs. Analyzing the results

of the literature review, inventory level assessed monetarily is the most employed metric

in papers. It is true that some expenses like depreciation and insurance could be included

in total warehouse costs and not necessarily in inventory costs. However, considering just

inventory level seems to be an incomplete way of measurement since other expenses like

holding cost and stock out penalty are also taken into account by other authors like Rimiene

(2008) and Li et al. (2009). So, the inventory cost denition used in this work follows Li

et al. (2009).

The nal group of cost indicators are presented in Equation 5.22 up to Equation 5.27,

with the meaning of data utilized in cost indicators described in Table 5.7.

TrC $

Trc = ( ) (5.23)

Ord Del order

5.2. Analytical model of Indicator Equations 89

Notation Denition

InvC = nancial cost to maintain inventory in warehouse per month($/month)

penalty measured by company as a cost when the customer makes an order and the

LostC =

product is not available per month ($/month)

total transportation cost, which is the sum of assets, oil, maintenance and labor

TrC =

costs per month ($/month)

Ord Del = number of orders delivered per month(nb/month)

Ord ProcC = sum of oce and employee costs to process orders per month ($/month)

Cust Ord = number of customer orders per month (nb/month)

WarC = sum of all activity costs that the warehouse has in charge per month ($/month)

Sales = total revenue from sales per month ($/month)

Salary = total salaries of all warehouse employees per month ($/month)

Charges = total charges paid over salary for all warehouse employees per month ($/month)

BuildC = total cost to maintain warehouse building per month($/month)

EqMaintC = total equipment maintenance costs per month ($/month)

Others = other costs not dened in the formulas per month ($/month)

Ord ProcC $

OrdProcc = ( order ) (5.24)

Cust Ord

WarC

CSc = × 100(%) (5.25)

Sales

$

Labc = Salary + Charges + Others( month ) (5.26)

$

Maintc = BuildC + EqMaintC + Others( month ) (5.27)

The quality indicators are presented in Equation 5.28 up to Equation 5.41, derived from

metric denitions (Table 5.8). These indicators measure characteristics of the products

and the work performed in a quantitative way. The indicator data are described in Table

5.9.

The distinction between the indicators on time delivery and orders shipped on time

(see Table 5.8) resides in what is considered as the nal monitoring point. On time delivery

words, if the warehouse monitors the delivery activity, it will use the indicator on time

delivery. The indicator orders shipped on time does not include the delivery activity

and if the warehouse measures their indicators up to the shipping activity (i.e. the moment

when the products leave the warehouse), it will use the indicator orders shipped on time. In

this work both measures are maintained in the metric system to evaluate their interaction

Cor Unlo

Recq = × 100(%) (5.28)

Pal Unlo

90 5. Conceptualization

Receiving pallets unloaded

Recq our denition (5.28)b

accuracy without incidents

Storage ac- storing products in Voss et al. (2005);

Stoq (5.29)a

curacy proper locations Rimiene (2008)

movement of the

right product from

Replenishment storage area to

Repq our denition (5.30)b

accuracy the right place in

forward pick area,

without damages

the physical counts

Physical in- of inventory agree

Bowersox et al.

Invq ventory ac- with the inventory (5.31)a

(2002)

curacy status reported in

the database

number of orders

Picking ac- Bowersox et al.

Pickq picked correctly per (5.32)a

curacy (2002)

orders picked

De Koster and

Orders

number of errors Waremius (2005);

Shipq shipped (5.33)a

free orders shipped De Koster and

accuracy

Balk (2008)

number of orders

Delivery ac-

Delq distributed without Campos (2004) (5.34)a

curacy

incidents

Voss et al. (2005);

number of orders

Forslund and Jons-

On time de- received on time

OTDelq son (2010); Lu and (5.35)

livery or before commit-

Yang (2010); Yang

ted date

and Chen (2012)

Orders number of orders

Kiefer and Novack

OTShipq shipped on shipped on time per (5.36)

(1999)

time total orders shipped

number of orders

Order ll Ramaa et al.

OrdFq lled completely on (5.37)

rate (2012)

the rst shipment

number of orders

delivered on time,

Perfect Kiefer and Novack

PerfOrdq without damage (5.38)

order (1999)

and with accurate

documentation

number of cus-

Lao et al. (2011);

Customer tomer complaints

CustSatq Voss et al. (2005); (5.39)

satisfaction per number of

Lao et al. (2012)

orders

Lao et al. (2011);

number of stock

Stockout Yang and Chen

StockOutq products out of (5.40)a

rate (2012); Lao et al.

order

(2012)

Rate of product loss

Scrapq Scrap rate Voss et al. (2005) (5.41)a

and damage

a

Interpretation of the indicator denition b This indicator is not explicitly dened

in the literature and we consider the denition presented in this table for the purpose

of this work.

5.2. Analytical model of Indicator Equations 91

Notation Denition

number of orders delivered complete in one shipment per month

Complet 1st Ship=

(orders/month)

Cor Unlo = number of pallets unloaded correctly per month (pallets/month)

Cor Sto = number of pallets stored correctly per month (pallets/month)

number of pallets moved correctly from reserve storage to forward picking area

Cor Rep =

per month (pallets/month)

Cor OrdLi Pick = number of order lines picked correctly per month (order lines/month)

Cor OrdLi Ship = number of order lines shipped correctly per month (order lines/month)

Cor Del = number of orders delivered correctly per month (orders/month)

number of orders with customer complaints regarding on logistics aspects per

Cust Complain=

month (orders/month)

number of products per month that are not available in stock when the customer

Prod noAvail=

makes an order (product/month)

Nb Scrap= number of scraps occurred in warehouse operations per month (product/month)

OrdLi Pick = number of order lines picked per month (order lines/month)

OrdLi Ship = number of order lines shipped per month (order lines/month)

Ord Ship = number of orders shipped per month (orders/month)

Ord Del = number of orders delivered per month (orders/month)

number of orders received by customer on or before deadline per month

Ord Del OT=

(orders/month)

Ord Ship OT= number of orders shipped on or before the deadline per month (orders/month)

number of orders received by customer on time (OT), with no damages (ND)

Ord OT, ND, CD=

and correct documentation (CD) per month (orders/month)

number of pallets with inaccuracies between the physical inventory and the

Prob data =

system per month (pallets/month)

Prod Out = number of products taken out of the inventory per month (product/month)

Pal Unlo = number of pallets unloaded per month (pallets/month)

Pal Sto = number of pallets stored per month (pallets/month)

number of pallets moved during replenishment operation per month

Pal Moved =

(pallets/month)

92 5. Conceptualization

Cor Sto

Stoq = × 100(%) (5.29)

Pal Sto

Cor Rep

Repq = × 100(%) (5.30)

Pal Moved

Invq = × 100(%) (5.31)

Pal Unlo + Pal Sto + Pal Moved

Pickq = × 100(%) (5.32)

OrdLi Pick

Shipq = × 100(%) (5.33)

OrdLi Ship

Cor Del

Delq = × 100(%) (5.34)

Ord Del

Ord Del OT

OTDelq = × 100(%) (5.35)

Ord Del

Ord Ship OT

OTShipq = × 100(%) (5.36)

Ord Ship

OrdFq = × 100(%) (5.37)

Ord Ship

PerfOrdq = × 100(%) (5.38)

Ord Del

CustSatq = × 100(%) (5.39)

Ord Del

Prod noAvail

StockOutq = × 100(%) (5.40)

Prod Out

Nb Scrap

Scrapq = × 100(%) (5.41)

Prod Proc

and Data

5.3.1 The Construction of Data Equations

The rst part of the analytical model encompasses the indicator equations presented in

the previous sections. To achieve the complete analytical model, we elaborate quantitative

expressions for indicator data to nd theoretically the indicator relationships (performed

in Section 6.3). The purpose of creating data equations is to verify their relationships,

identifying the independent and combined data. The combined data is measure from other

data, e.g. data J in Equation 4.2, Chapter 4, is a combined data since it is calculated from

5.3. Complete Analytical Model of Performance Indicators and Data 93

the sum of A and G data. The independent data are the real inputs of the system, i.e. they

are not calculated from any other data (e.g. A and G in Equation 4.2 are independents).

The complete analytical model, presented in Appendix A, has one more data format

besides the ones already presented (indicator's name are in bold, as Rect , and data used

in indicator equations are in sans serif style, e.g. Pal Unlo): the components inside data

equation are in slanted style like Prob Rep. In the cases where the same component is used

in indicator equation and in data equation, we choose to format it in the higher level.

For instance, the term OrdLi Ship is used as indicator data in Equation 5.14 and also as

Labp = ( hour ) (5.9)

WH

Shipp = ( hour ) (5.14)

WH Ship

$

Labc = Salary + Charges + Others( month ) (5.26)

Initially, analyzing these equations, one could infer that all these 7 dierent data are

independents because it is not possible to calculate one in terms of another. However, there

are just two independent data: OrdLi Ship and Others. The term Others is independent

and has no relationship with any data presented. The other six data form two dierent

groups: Prod Proc is calculated from OrdLi Ship, and WH, WH Ship, Salary and Charges

have relationships. The relationships among data (and consequently among indicators) are

developed from the data equations, presented in Equations 5.42, 5.43, 5.44, 5.45, 5.46.

Equation 5.42 is developed from the data denition described in Table 5.5, where

where Prod Proc is the number of products processed in the warehouse, represented by the

shipped products, OrdLi Ship are order lines shipped, Prod Ship is the number of products

shipped and Prod Line is the average number of products in a shipping order line. From

this equation we conclude that OrdLi Ship and Prod Line are independent data.

Analyzing the data equations of the other four data (WH, WH Ship, Salary and Charges)

we have:

where WH is the total available working hours for all warehouse activities (WH Rec,

WH Sto, WH Rep, WH Pick, WH Ship, WH Others). The available working hours for a

specic activity (e.g. WH Ship, used in indicator Equation 5.14) is calculated as the aver-

age number of employees working in storing (nb of employees) times the total number of

94 5. Conceptualization

+ $/hother × WH Others (5.45)

where Salary encompasses the total amount payed for all shop oor employees of each

activity. $/h is the remuneration value per hour for each activity ($/hrec , $/hsto , $/hrep ,

$/hpick , $/hship ). βord is an index to represent the percentage of the total available labor

hours the employees are dedicated to customer orders administration. These customer

orders working hours are included in OrdProcc indicator (Equation 5.24), and the working

hours left is considered in Salary equation. α is an index to represent the partial quantity

It is possible to see from Equations 5.43 - 5.46 that WH, WH Ship, Salary and Charges

are combined data since they are computed from other informations. The real inputs from

these equations are: $/h of all activities, nb of employees of each activity, WarWH, βord

and α.

Therefore, Equation 5.9, 5.14 and 5.26 have as real inputs to be calculated (independent

data): OrdLi Ship, Prod Line, $/h of all activities, nb of employees of each activity, WarWH,

βord and α, Others.

As demonstrated here through an example, we have elaborated expressions for data of

the 41 indicators. The complete analytical model derived from data and indicator equations

is exhibited in Appendix A.

The analytical model should be developed according to the context of the studied ware-

house, since the specicities of each warehouse result in dierent equations. Therefore,

the developed analytical model refers to the standard warehouse presented throughout this

chapter.

Even if it is not possible to generalize the analytical model, the proposed equations

can help with the development of analytical models in other warehouse contexts. For

this reason, the term others is included in some equations to allow their adjustments if

necessary.

The main assumptions made which impact equation denitions are as follows:

• The inventory cost is not a part of the total warehouse costs. The reason is that

inventory costs are usually a charge of the enterprise as a whole, and the warehouse

5.4. Conclusions 95

• The distribution cost (Equation 5.23) does not make part of total warehouse costs

As the costs incurred for delivery activity usually have no relation with the internal

• Trucks used for delivery are enterprise's assets. Therefore, distribution costs includes

truck maintenance. If the company has an outsourced distribution, all these compo-

nents are changed by the monthly value paid for the third party logistics company

house costs;

• The quality data is dened as a sum of a process made correctly and with problems,

and this division allows the identication of quality problems through the process.

Assume the delivery accuracy (Equation 5.34), which is measured by orders delivered

correctly per total orders delivered. In the total orders delivered, a portion of it may

be delivered correctly while the other part may not. This other part is named orders

delivered with problems, Prob Del. But it does not mean that the order could not be

delivered, it just means that this order is recorded with quality issues. For example,

the number of orders not delivered on time are counted in Prob Del even if they

• Two data are dierentiated even if their results can occasionally be the same. For

example, the number of orders delivered, Ord Del, is not considered as equal to cus-

tomer orders Cust Ord. Even if these numbers will be close to each other, usually

there are orders in process inside the warehouse at the end of the month, when the

data is collected to measure indicators. Some orders have already been processed by

the administration but not delivered yet. Thus, to calculate the order processing cost

indicator, OrdProcc , the total customer orders are taken into account while for the

order lead time indicator, OrdLTt , orders delivered are considered. Other similar

5.4 Conclusions

The main objective of this chapter is to develop an analytical model of performance indi-

This chapter starts with the presentation of the theoretical warehouse studied (named

Standard Warehouse) with its layout and activities. After, the metric system to assess

warehouse performance is dened, rstly based on the literature review. A total of 41 indi-

cators compose the metric system, representing all activities that the standard warehouse

has in charge.

In order to create the analytical model, the indicator denitions are rst interpreted

in order to build indicator equations. From these results, data equations are developed,

expanding indicator equations and providing information about the kind of inputs used in

the analytical model. The complete analytical model demonstrates all relations among data

and in the next chapter it will be used to determine indicator relationships theoretically.

Chapter 6

Modeling

important, abused and misused.

Sink, 1991.

Contents

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

6.2 Data generation for the Standard Warehouse . . . . . . . . . . 98

6.2.1 Assumptions in data generation . . . . . . . . . . . . . . . . . . . . . . . 98

6.2.2 The global warehouse scenario . . . . . . . . . . . . . . . . . . . . . . . . 99

6.2.3 The internal warehouse scenario . . . . . . . . . . . . . . . . . . . . . . . 100

6.2.4 Data characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

6.3 Theoretical model of Indicator relationships . . . . . . . . . . . 102

6.3.1 The data associations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

6.3.2 Determination of the independent data . . . . . . . . . . . . . . . . . . . 103

6.3.3 Data versus indicator relationships . . . . . . . . . . . . . . . . . . . . . 105

6.3.4 Analysis of indicator relationships . . . . . . . . . . . . . . . . . . . . . . 107

6.4 Statistical Tools Application . . . . . . . . . . . . . . . . . . . . . . 111

6.4.1 Data normality test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

6.4.2 Correlation measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

6.4.3 Principal Component Analysis . . . . . . . . . . . . . . . . . . . . . . . 115

6.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

Abstract

In this chapter, a scenario representing the ow of products between processes for the

standard warehouse is designed. This scenario is used to generate shop-oor monthly

data, which are utilized to measure performance indicators. The dataset formed by

performance indicators measured monthly are the inputs of the mathematical tools used

to model indicator relationships. Firstly, the Jacobian matrix is assessed and the results

give some insights about the relationships between indicators based on their equations.

Secondly, statistical tools are applied to propose a model for indicators' aggregation.

The rst results suggest that the relationships between indicators are mainly based on

their measurement domain, i.e. the indicators are aggregated according to warehouse

activities.

98 6. Modeling

6.1 Introduction

This chapter performs the modeling phase of the methodology. The main objectives of

this phase are to provide the theoretical model of indicator relationships and to apply the

statistical tool, obtaining the rst insights about the indicators' aggregation.

To reach these objectives, a dataset is necessary. In a real context, data from the

warehouse shop oor exists in databases or can be collected. However, as our studied

warehouse is theoretical, we generate data for the standard warehouse, representing its

ow of products between processes. This initial dataset is used to calculate performance

indicators monthly, creating indicator time series that are coupled with the mathematical

tools.

from the assessment of the Jacobian matrix. Finally, some statistical tools are performed

6.2.1 Assumptions in data generation

The main scenario created for data generation occurs in the shop-oor of the standard

preferred to generate the data used to calculate indicators, which are the ones presented in

the analytical model of Chapter 5. The reason for this choice is that there is great quantity

of relationships among all data which directly impact indicator results (for instance, the

same data can be used to calculate another data and some indicators, see the example in

Section 5.3.1). If the indicator results are generated directly, these relationships may be

lost (e.g. it may not be possible to see the impact of a data change in the indicator results).

Thus, it could become more dicult to group indicators according to their relationships.

There are a lot of methods for data generation. In this work, a spreadsheet in Excelr

software is developed. The Excel

r spreadsheet is elaborated to create data following

normal and random functions and to represent the eect of chained processes (as discussed

in Chapter 4, Section 4.3.2). Due to the diculty of representing reality, some assumptions

zero;

• The supplier orders have always the same quantity, a truck of 10 tons with 25 pallets;

• An order can not present two dierent errors within the same month;

• The warehouse processes only one product and it is possible to put 40 products in a

pallet;

6.2. Data generation for the Standard Warehouse 99

• The indicators Perfect order, PerfOrdq , and Delivery quality, Delq , (Equations 5.38

and 5.34, Section 5.2.7 ) consider that an order is perfect (and, consequently, correct)

if it is on time, with no damages, with the right quantity and the right documents.

Due to this consideration, the number of correct orders delivered (Cor Del) and the

number of perfect orders (Ord OT, ND, CD) are equal, resulting that both indicators

remain the same equation. Thus, delivery quality is eliminated from the metric set,

It is important to discuss the assumption that the warehouse manages and delivers

just one product. Even if it seems a restrictive assumption, the data created with one or

several products does not change substantially indicator results, which are calculated with

the data generated. The two following examples demonstrate the impacts of this decision

The inbound operations usually use the unit `pallet' to measure indicators. In some

cases (e.g. the indicator Labor productivity, Labp ), it is also necessary to know the number

of products that are in the pallets. Even if there are dierent products in a pallet, the

interest is in the total number of products received in pallets, which will not change for

one or several kinds of products. Consequently, the scenario considering one product does

not modify the nal indicator results for the inbound operations.

For the outbound operations, the assumption that an order has just one kind of product

results in `number of orders' and `number of order lines' with the same quantity, since all

orders have just one line. However, this situation impacts only the productivity and time

indicators for picking and shipping activities (total of 4 indicators) from the 40 indicators

Therefore, we consider that these data can be used to represent a warehouse operation

Figure 6.1 shows the global scenario of the standard warehouse. The main informations

assume that the warehouse has 5.000 m2 of area, operates eight hours per day and can

store 1000 pallets. The information about the proportion of pallets capacity in a warehouse

area of 5.000 m2 is acquired from specialized websites about warehouse construction (e.g.

www.spartanwarehouse.com/warehouse-space-calculator).

From these characteristics, the quantity of products entering and leaving the warehouse

every month is, on average, 28000 units. Figure 6.1 depicts products arriving in trucks of

10 tons (with 25 pallets per truck and 40 products per pallet) and orders leaving the

The objective of creating this scenario is to measure the warehouse performance for all

activities, i.e. from the product arrival at the dock to be unloaded up to order delivery to

the client. For indicators' measurement, we assume that this warehouse collects data once

a month, commonly in the last working day, and these data signify all eorts made during

100 6. Modeling

Inbound: Outbound:

Average of 28 procurements / month Truck capacity = 5 tons = 12 pallets

1 pallet = 40 products Delivery travels per truck = 3 / day

Truck capacity = 10 tons = 25 pallets Average of 20 products per order

Receive in average 28000 products/month Average of 28000 products processed/ month

Warehouse:

Warehouse area = 5000 m2

Inventory capacity = 1000 pallets

Warehouse working hours = 8h/day

Figure 6.1: Main informations about warehouse, inbound and outbound activities.

the month to process supplier and customer orders. Hence, the data generated represents

a summary of all that has been processed by the warehouse during the month.

The detailed warehouse scenario is shown in Figure 6.2, representing a picture of the

warehouse activities at the end of the last working day of the month. This gure represents

the product and information ows occurred during the month; these data are obtained to

There are three kinds of symbols in Figure 6.2:

ow of products inside the warehouse with their associated information; · · · m shows the

information ow in an activity or between warehouse areas; · · · •is the internal data inputs

(IntInput) and outputs (IntOutput) used to measure indicators related to a specic activity.

The notation used in the inputs and outputs of activities is the same as the ones presented

Supplier Ord Cor Pick + Prob Pick Cust Ord

+

+ Cor Sto + Cor Rep +

Cor Unlo + + + Cor Pick + + Cor Ship +

+ + + + + + Cor Del +

+ + + +

Prob Sto Prob Rep

Prob Unlo Prob Pick Prob Ship +

No Proc + No Proc +

+ + -

IntInput IntOutput

IntInput IntOutput IntInput IntOutput IntInput IntInput IntOutput IntInput IntOutput

IntOutput

Scrap Unlo1 Sto inProcess

Scrap Del1

LEGEND:

Product flow with its information Reserve Stock area

Information flow Forward Picking area

Local data No Proc – Products not processed

Figure 6.2: Flow of products and information throughout the warehouse activities.

In Figure 6.2, the product ows throughout warehouse areas are demonstrated by the

6.2. Data generation for the Standard Warehouse 101

The inputs vary among activities. The storage, shipping and delivery have as main

inputs the products processed by the previous activity, and the receiving depends on the

number of supplier orders requested in the month. For replenishment, the products are

taken out from the reserve stock area according to the total number of orders picked

(Cor Pick + Prob Pick), since the forward picking area needs to have space to receive the

replenishment. Finally, the picking activity takes products out of the forward picking area

The outputs for all activities are the total of products processed correctly and with

problem (for instance, in storage activity, the outputs are correct pallets stored,Cor Sto,

and pallets stored with problems, Prob Sto). The outputs with problems are divided into

two categories: the problems totally solved during the month, allowing the products to

advance to the next process (as demonstrated by the arrow added to correct products);

and the problems that have not been solved yet, which are added in the next month to

the number of products that should be processed (information arrow added to `No Proc').

Therefore, the `problems not solved' impacts the product ows (e.g. scrap) while the

others (considered as solved) are just registered for indicators' measurement but they do

not impede product ows (e.g. data information error). As the solved problems make

part of products processed, the two outputs (activity performed correctly and with solved

Some activities have, at the end of the month, products that are not processed yet

(dened as `No Proc' in Figure 6.2). It means that not all supplier and customer orders

received in the month have already been completely processed. The sum of products with

problems not solved (dened above) and products not processed result in the products in

Process (e.g. Sto inProcess, Figure 6.2). These products in Process are not considered in

performance measures but they are included as inputs to the activity of the next month.

For simplication, the receiving and delivery activities do not have `No Proc' products.

As demonstrated in Figure 6.2, these activities do not have products not processed, which

means that there are no more trucks to unload (in receiving) and all pallets loaded during

the day were delivered (in delivery). For both activities, just the products with non solved

From this scenario, data is built for each warehouse activity, as shown in Appendix

B. The next section summarizes the dierent kinds of data generated and presents some

examples.

As stated earlier, a spreadsheet is designed to represent the activities described in Figure

6.2. Due to the complexity of the warehouse scenario, dierent categories of data are

necessary to better represent process variabilities. They are distinguished as xed, uniform,

The xed data are established values that will not change over time, e.g. warehouse

space, number of equipments, warehouse opening hours per day, number of employees.

The `uniform data' is a random number generated from a uniform distribution of prob-

abilities with pre-dened limits (function `randbetween ' in Excelr ). These limits can be

xed (for instance, the number of days per month that the warehouse operates varies be-

102 6. Modeling

tween 20 and 25) or variable (if the limits are determined by other variables). As an example

of this last case, the number of products stored correctly can not be higher than the total

number of products processed in receiving activity. Hence, the number of products stored

will have its limits dened by the outputs of the receiving process. These kinds of limits

are applied for all the warehouse activities, representing the eect of chained processes.

Finally, the normal function calculates a certain probability using the normal distribu-

tion according to a given mean and standard deviation (function `norminv' in Excelr ).

This function is utilized in dierent situations along the warehouse data generation. For

instance, the range of products received and delivered in a month follows a normal distri-

bution, with mean of 28000 products and standard deviation of 2000 products. Moreover,

the number of products per order uses the same function with mean of 20 products and

standard deviation of 2.

The complete list is presented in Appendix B, where all equations used to generate

the products processed in the warehouse during a whole month. These indicators assessed

monthly are used as inputs of the mathematical tools to nd indicator relationships. In the

next section, the theoretical model introduced in Chapter 4, Section 4.3.2 is implemented

The complete analytical model dened in Chapter 5 demonstrates that the relations among

data are complex, making the global performance hard to evaluate taking into account the

data dependency. So, it is crucial to understand these relationships to better evaluate the

warehouse performance.

Section 4.3.2 has presented how to verify indicator relationships analyzing indicator

equations. In this section we perform this analysis for the complete analytical model of

40 indicators with their data equations. Initially, we have carried out a manual procedure

achieved are not exhaustive; not all data relationships are taken into account. Thus, we

First of all, the data equations from the complete analytical model (see Appendix A)

are studied to dierentiate the independent data from the combined data (as dened in

Section 5.3.1, the combined data is measured from other data, whereas the independent

ones are the real inputs of the system). Once the independent data are identied, we verify

the total number of indicators related by one or more data inputs. For that we use the

partial derivative matrix of indicator equations. Finally, the indicator relationships are

discussed.

6.3. Theoretical model of Indicator relationships 103

To get the results of this exhaustive procedure, we utilize the software CADES

r (Com-

1. CADES has three main

The rst module, CADES Generator, allows to code the analytical model of equations in

sml language (System Modeling Language) (Enciu et al., 2010). The model equations that

can be implemented in sml are analytical and/or semi-analytical. When CADES compiles a

model written in sml, it calculates automatically its gradient by using derivation techniques

and the result is an icar component containing the model output functions in terms of

the inputs (Staudt, 2015). The Jacobian matrix of the system is calculated in CADES

Calculator, the second module, using the exact derivatives obtained in CADES Generator.

Finally, the third module, CADES Optimizer, allows to couple the icar component directly

to optimization algorithms (more details of this module are presented in Section 7.4.3).

In Section 5.3.1, an example was carried out to demonstrate how data are highly connected,

with some data making part of more general ones. Regarding this situation, Figure 6.3

depicts the combined data with their main elements for the majority of the indicator set.

The rectangle colors do not have a special meaning; Figure 6.3 demonstrates data in the

external rectangles comprehending the data from the internal ones. For example, Equation

A.3 shows that unloaded pallets can be divided into pallets unloaded correctly and with

problems. The rst rectangle in the upper left side of Figure 6.3 represents this equation.

The blue rectangle concerns all pallets unloaded (sum of data) and inside it there are two

other rectangles corresponding to the pallets unloaded correctly Cor Unlo and the pallets

unloaded with problems Prob Unlo . Yet, Prob Unlo have two other data represented

by the rectangles 1 and 2, signifying, respectively, the scraps and data system errors

Figure 6.3 is divided in four areas: inbound, outbound, resource and general. The in-

bound and outbound contain data regarding the activities executed in this warehouse areas.

The resource data is related to capacity and the general data concern several warehouse

activities; that is the reason why they are separated from the other data.

We can infer from Figure 6.3 that it is hard to identify the independent data with so

many relations among them. Thus, next section determines the independent data using

the CADES

r software.

After the identication of data association, we want to obtain a list of the independent data

necessary to assess the 40 indicator set. Due to the big quantity of information to evaluate

(all equations of the complete analytical model in Appendix A), it is dicult to make

manually the same analysis performed in Section 5.3.1, in order to dene the independent

and combined inputs of the system. Therefore, the complete analytical model (without

the data components named Others) is coupled with the software CADES Generator to

obtain all the inputs (independent data) and outputs (indicators) of the equations.

1

http://www.vesta-system.fr/fr/produits/cades/

104 6. Modeling

INBOUND DATA

Cor 1 Prob 2

WH

Unlo Unlo WEfRec

Rec

Pallets Unloaded ∆T

∆TRec 5

Dock

to

Cor 1 Prob 2 WH

WEfSto Stock

Sto Sto Sto

∆TSto 5

Pallets Stored

OUTBOUND DATA WH

WEfRep

1 2

Rep

Cor Prob

Rep Rep ∆TRep 5

Pallets Replenished WH

WEfPick

Pick

1 Prob OrdLi

Cor OrdLi ∆TPick 5

Salary &

Pick 4 Pick

Charges

Prod noAvail

WH ∆T

OrderLi Picked WH WEfShip

Ship Order

∆TShip 5 Lead

Cor OrdLi 1 4 Labor Cost Time

Ship Prob OrdLi WH

Complet Ship Transport Cost WEfDel

Del

1st Ship Deprec Cost Salary

∆TDel 5

OrdLi Ship / Prod Ship Truck Maint

Oil Cost Cost

Orders Shipped

Docs OK On Time Del

1 4

No Damage

Orders Delivered / Kg Tr Kg Available

Customer Orders

1 Scrap 2

War Capacity Error

HEq HEq data system

War Cap Stop Work

Used 4 Data

HEq Avail Sales

Inv Cap error

Used Prod Prod

WH 5 WEf Profit Cost

Inv Cap Admin Admin

6.3. Theoretical model of Indicator relationships 105

The compilation provides as outputs the 40 indicators studied in this work and the

input's list contains 81 independent data, as shown in Table 6.1. The meaning of each

Data Inputs

α deprec2 kg Prod Profit

β_del empl Admin l_used Rate

β_ord empl Del mean_Insp Remain_Inv

β_pick empl Pick nbMachine scrap1

β_rec empl Rec nb_travel scrap2

β_rep empl Rep NoComplet Ord Ship scrap3

β_ship empl Ship Ord Del OT scrap4

β_sto empl Sto Ord Ship OT scrap5

BuildC EqMaintC pal_truck scrap6

cap error data system1 pallet_area Truck Maint C

Cor OrdLi Pick error data system2 Prob OrdLi Pick War Cap

Cor OrdLi Ship error data system3 Prob OrdLi Ship war used area

Cor Del HAdmindel Prob Del War WH

Cor Rep HAdminpick Prob Rep $/hadmin

Cor Sto HAdminrec Prob Sto $/hdel

Cor Unlo HAdminrep Prob Unlo $/hpick

Cust Ord HAdminship Prod Ord $/hrec

Cust Complain HAdminsto Prod pal $/hrep

ΔT(Insp)2 HEq Stop Prod noAvail $/hship

deprec1 Inv Cap Prod Cost $/hsto

$ oil

After the determination of this nal data list, we proceed with the verication of the

indicator relationships.

To check all indicators that have relations by the use of the same data we use the Jacobian

In our case, we derive all functions f (indicator equations, from Equation 5.1 to Equa-

tion 5.41, excepting Delq ) with respect to their data inputs x (presented above, Table

6.1). So, the nal partial derivative matrix has the size 40 × 81 (n × m), where n are the

indicators and m the data inputs.

Due to the substantial size of the partial derivative matrix, we also automatize the

Before getting the results of the Jacobian matrix, it is necessary to provide initial values

to the inputs. The assigned values correspond to the rst month of the data generated for

our warehouse scenario, presented in the beginning of this chapter (see Appendix D for

r computes and gives the

numerical results of the Jacobian matrix for the supplied input data set. Figure 6.4 shows

the software interface with the inputs, outputs, and the Jacobian matrix result.

106 6. Modeling

INPUTS: 81

INDEPENDENT DATA

OUTPUTS: ALL 41

INDICATORS

JACOBIAN MATRIX:

41 x 81

r software: inputs, outputs and Jacobian matrix areas.

The calculated Jacobian matrix is initially analyzed with respect to its columns. We

observe that there are two main kinds of inputs (columns of the matrix): the ones related

to only one output (see Table 6.2) and the others linked to several outputs (see Table

6.3). For illustration purposes, only some parts of the matrix are shown in Tables 6.2 and

6.3. Each cell, in both tables, contains the partial derivative values of the output with

respect to the corresponding input data. The partial derivative value can be interpreted as

the variation of the output when the corresponding input varies, maintaining other inputs

constant.

From the 81 data inputs, 27 are associated with one output and 54 with two or more

outputs. In Tables 6.2 and 6.3, the most signicant values inuencing positively and

negatively the indicators are highlighted in red and green colors, respectively.

Table 6.3 presents the basis used to determine indicator relationships. The assumed

preliminary hypothesis of this thesis mentions that two indicators with non-zero partial

derivative for the same input might have a relationship between them. Evaluating two

6.3. Theoretical model of Indicator relationships 107

Table 6.2: Partial area of Jacobian matrix with inputs related to just one output.

NoComplet_

Cust_Ord CustComplain ErrorDataSystem3 nbMachine OTDel_ord OTShip_ord palSpace

OrdShip

CSc 0 0 0 0 0 0 0 0

CustSatq 0 -0,07496 0 0 0 0 0 0

Delp 0 0 0 0 0 0 0 0

Delt 0 0 0 0 0 0 0 0

DSt 0 0 0 0 0 0 0 0

EqDp 0 0 0 -2,1 0 0 0 0

Invc 0 0 0 0 0 0 0 0

Invq 0 0 -0,05006 0 0 0 0 0

InvUtp 0 0 0 0 0 0 0 -0,07192

Labc 0 0 0 0 0 0 0 0

Labp 0 0 0 0 0 0 0 0

Maintc 0 0 0 0 0 0 0 0

OrdFq 0 0 0 0 -0,07402 0 0 0

OrdLTt 0 0 0 0 0 0 0 0

OrdProcc -0,00155 0 0 0 0 0 0 0

OTDelq 0 0 0 0 0 0,07496 0 0

OTShipq 0 0 0 0 0 0 0,07402 0

PerfOrdq 0 0 0 0 0 0 0 0

Table 6.3: A partial view of the Jacobian matrix with inputs related to two or more outputs.

alpha beta_del beta_ord CorDel CorRep CorSto CorUnlo emplPick emplRec emplRep emplShip emplSto

CSc 0,24550 0 0,00000 -0,00038 0 0 0 0,02593 0,02593 0,02593 0,02593 0,02593

CustSatq 0 0 0 0,00101 0 0 0 0 0 0 0 0

Delp 0 0 0 0,00298 0 0 0 0 0 0 0 0

Delt 0 0,25190 0 -0,00021 0 0 0 0 0 0 0 0

DSt 0 0 0 0 0 0 -0,00067 0 0,20400 0 0 0,20400

EqDp 0 0 0 0 0 0 0 0 0 0 0 0

Invc 0 0 0 0 0 199,8 0 0 0 0 0 0

Invq 0 0 0 0 0,00013 0,00013 0,00013 0 0 0 0 0

InvUtp 0 0 0 0 0 0,05000 0 0 0 0 0 0

Labc 9988,0 0 -5292,0 0 0 0 0 1260,0 1260,0 1260,0 1260,0 1260,0

Labp 0 0 0 0 0 0 0 -1,5 -1,5 -1,5 -1,5 -1,5

Maintc 0 0 0 0 0 0 0 0 0 0 0 0

OrdFq 0 0 0 0 0 0 0 0 0 0 0 0

OrdLTt 0 0,25190 0,37780 -0,00101 0 0 0 0,11960 0 0 0,11960 0

OrdProcc 1,4 0 3,7 0 0 0 0 0 0 0 0 0

OTDelq 0 0 0 -0,07367 0 0 0 0 0 0 0 0

Putt 0 0 0 0 0 -0,00036 0 0 0 0 0 0,21120

Recp 0 0 0 0 0 0 0,00595 0 -4,2 0 0 0

Recq 0 0 0 0 0 0 0,00184 0 0 0 0 0

Rect 0 0 0 0 0 0 -0,00033 0 0,20400 0 0 0

Repp 0 0 0 0 0,00595 0 0 0 0 -3,7 0 0

dierent rows of Table 6.3 (i.e. two indicators), we observe several common inputs. This

is the case of Labc and OrdLTt , which have in common three inputs: βord , emplPick,

emplShip, denoting a relationship between them. Therefore, after comparing two rows of

Table 6.3 each time, we check all possible relations among indicators.

The results presented by the Jacobian matrix (Table 6.3) are analyzed in terms of: the

number of data shared by two indicators; the numerical values of the partial derivatives.

The main objective of both analysis is to try to gure out the intensity of indicator rela-

tionships.

Table 6.4 shows the number of shared data between indicators for the complete Jacobian

matrix, and the colors represent: red for 1 shared data, blue for 2, and green cells represent

108 6. Modeling

3 or more shared data. From this table, three main results are interesting to discuss:

The white cells with zero values represent that indicators have no data in common,

making easy the interpretation. Indicators that do not share data with others should have

no relationships, and consequently, may not make part of the indicator group which will

In the opposite of the white cells, the green ones show indicators sharing three or more

data. One may deduce that the greater number of shared data determine higher indicator

relationships. Taking the rst column, of CSc indicator, it is possible to see that it shares

data with 11 other indicators. The three most expressive numbers of shared data are

15 with Labc and 7 with Labp and OrdLTt . From this result we may conclude that

these indicators have high relationships, specially between CSc and Labc . However, the

correlation between CSc and Labc is only 0,55, a medium value, whereas between CSc

and Labp is -0,96, denoting a very high correlation (Section 6.4.2 presents the complete

correlation matrix). Therefore, the hypothesis that a great number of shared data signies

Due to the conclusion for indicators with several data in common, the indicators with

few data (the red and blue cells of Table 6.4, which are the majority of situations) are even

It seems that other situation that impact the nal relationship between indicators is

the numerical values of the partial derivatives. Analyzing the column beta_ord of Table

6.3, the rows for Labc and OrdProcc demonstrate expressive values of partial derivative

(-5292 and 3,7 respectively) what might suggest the intensity of relationships. However,

as it can be noticed in Table 6.2 and 6.3, the numerical values of the partial derivatives

may dier substantially from one to another. At this time, it is interesting to recall that

the input data may have dierent units and their values can be in a distinct scale. For

example, the input number of employees, can be often a small number compared to the

average number of products in inventory, which is usually a big quantity. Moreover, the

Jacobian matrix is calculated by considering the monthly input data set. It means that for

each month the Jacobian matrix can slightly change, depending on the actual variation of

the inputs parameters. Due to the dynamic nature of the input data and also the numerical

dierence they might have (due to their units), it is hard to directly dene the intensity of

Therefore, it is not possible to infer about the intensity of indicator relationships from

the results obtained. The use of Jacobian matrix to dene the strength of indicator relation-

ships requires a deeper study, which is proposed as a future research direction. Regarding

this thesis, we utilize the results of the Jacobian matrix to give a preliminary overview of

indicator relationships in a qualitative sense, and to give support in the choice of the nal

From the exhaustive relationship matrix presented in Table 6.4, it is possible to create

the same framework as presented in Appendix C, Figure C.3. However, due to the great

6.3. Theoretical model of Indicator relationships 109

quantity of indicator relations, the result is not easily interpretable as in Figure C.3. For

that reason, this nal framework is placed in Appendix E just for illustration.

110

Table 6.4: Indicator relation matrix with the number of shared data.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40

1 CSc 0

2 CustSatq 2 0

3 Delp 3 2 0

4 Delt 3 2 4 0

5 DSt 3 0 1 1 0

6 EqDp 1 0 1 1 1 0

7 Invc 3 0 0 0 0 0 0

8 Invq 0 0 0 0 2 0 2 0

9 InvUtp 0 0 0 0 0 0 4 2 0

10 Labc 15 0 1 1 3 1 0 0 0 0

11 Labp 7 0 1 1 3 1 1 0 0 6 0

12 Maintc 2 0 0 0 0 0 0 0 0 0 0 0

13 OrdFq 0 0 0 0 0 0 0 0 0 0 2 0 0

14 OrdLTt 7 2 4 6 1 1 0 0 0 5 3 0 0 0

15 OrdProcc 6 0 1 1 1 1 0 0 0 5 1 0 0 3 0

16 OTDelq 2 2 2 2 0 0 0 0 0 0 0 0 0 2 0 0

17 OTShipq 0 0 0 0 0 0 0 0 0 0 2 0 2 0 0 0 0

18 PerfOrdq 2 2 2 2 0 0 0 0 0 0 0 0 0 2 0 2 0 0

19 Pickp 2 0 1 1 1 1 2 0 0 2 2 0 0 2 1 0 0 0 0

20 Pickq 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 2 0

21 Pickt 2 0 1 1 1 1 2 0 0 2 2 0 0 4 1 0 0 0 4 2 0

22 Putt 2 0 1 1 4 1 2 2 2 2 2 0 0 1 1 0 0 0 1 0 1 0

23 Recp 2 0 1 1 4 1 0 2 0 2 2 0 0 1 1 0 0 0 1 0 1 1 0

24 Recq 0 0 0 0 2 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0

25 Rect 2 0 1 1 8 1 0 2 0 2 2 0 0 1 1 0 0 0 1 0 1 1 4 2 0

26 Repp 2 0 1 1 1 1 0 2 0 2 2 0 0 1 1 0 0 0 1 0 1 1 1 0 1 0

27 Repq 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0

28 Rept 2 0 1 1 1 1 0 2 0 2 2 0 0 1 1 0 0 0 1 0 1 1 1 0 1 4 2 0

29 Scrapq 1 0 0 0 0 0 2 0 1 0 4 0 2 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0

30 Shipp 2 0 1 1 1 1 0 0 0 2 4 0 2 2 1 0 2 0 1 0 1 1 1 0 1 1 0 1 2 0

31 Shipq 0 0 0 0 0 0 0 0 0 0 2 0 2 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 2 2 0

6. Modeling

32 Shipt 2 0 1 1 1 1 0 0 0 2 4 0 2 5 1 0 2 0 1 0 1 1 1 0 1 1 0 1 2 4 2 0

33 StockOutq 1 0 0 0 0 0 5 0 0 0 1 0 0 0 0 0 0 0 2 2 2 0 0 0 0 0 0 0 1 0 0 0 0

34 Stop 2 0 1 1 2 1 2 2 2 2 2 0 0 1 1 0 0 0 1 0 1 4 1 0 1 1 0 1 0 1 0 1 0 0

35 Stoq 0 0 0 0 0 0 2 2 2 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 2 0

36 Thp 2 0 1 1 1 1 1 0 0 1 5 0 2 1 1 0 2 0 1 0 1 1 1 0 1 1 0 1 4 3 2 3 1 1 0 0

37 TOp 4 2 2 2 0 0 5 2 4 0 1 0 0 2 0 2 0 2 0 0 0 2 0 0 0 0 0 0 2 0 0 0 1 2 2 1 0

38 Trc 4 2 4 4 1 1 0 0 0 2 1 0 0 4 2 2 0 2 1 0 1 1 1 0 1 1 0 1 0 1 0 1 0 1 0 1 2 0

39 TrUtp 4 2 2 2 0 0 1 0 0 0 1 0 0 2 0 2 0 2 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 1 4 3 0

40 WarUtp 0 0 0 0 0 0 4 2 4 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 1 0 0 0 0 2 2 0 4 0 0 0

6.4. Statistical Tools Application 111

This section presents the application of statistical tools to analyze indicator relationships,

The data matrix used to perform statistical tools is 100 x 40 where the rows present

the dierent values taken by these indicators over time and the columns represent the

dierent indicators. Each cell contains the indicator value for a specic month. The choice

of generating data for 100 months comes from the requirements to apply the PCA tool,

which species that the sample must be bigger than the number of variables. Using this

database generated for 100 months, we rst perform a normality test, to describe the

Xactual − Xmean

Xnew = (4.3)

σX

where Xnew is the new value of the variable, Xactual is the real variable value, Xmean is

the time series mean of the variable dataset, σX is the standard deviation of the variable

time series.

Once the standardization is done, the indicator correlations are measured and the

principal component analysis is performed, completing the group of informations that will

However, the best results obtained exclude a great quantity of indicators from the model

(from the initial 40 indicators, only 11 remains after performing DFA). As our objective is

to maintain the majority of indicators to evaluate the global performance, we do not use

this result in our integrated model. We suggest further researches to apply dynamic factor

analysis with this purpose. The initial results obtained are reported in Appendix F.

The objectives of testing data normality are to know data characteristics and to verify if

there are outliers in the dataset. The data characteristics are sometimes useful to justify

the results obtained specially in the utilization of the data in statistical tools. In the case of

outliers, according to Section 3.3.2, PCA is sensitive to great dierences among variables.

Even if the data is normalized before PCA application, it is important to identify the

existence of outliers. For this purpose, the skewness and kurtosis are measured for the

variables. As stated is Section 4.3.1, if the skewness is higher than 2 or the is kurtosis is

higher than 7 a special analysis of the time series should be made (Newsom, 2015). If these

limits are exceeded, it is necessary to look for outliers in the time series, xing the wrong

r to accomplish the

Anderson-Darling test for each indicator time series. Moreover, the skewness and kurtosis

The Anderson-Darling test measures how well the data follow a particular distribu-

tion, considering in the null hypothesis that data follow a normal distribution. The null

hypotheses is rejected if p-value is smaller than a chosen alpha (usually 0.05 for 95% of

112 6. Modeling

condence and 0.01 for 99% of condence). We chose to reject the null hypothesis (i.e. to

consider that data has a not-normal distribution) for p-values < 0.01.

Figure 6.5 presents some examples of these tests (Anderson-Darling, skewness and

kurtosis) for the indicators: Cost as a % of Sales (CSc ), Labor costs (Labc ), Inventory

quality (Invq ). The results are highlighted by red rectangles in the gure. For the skewness

and kurtosis, none of the results are greater than 2 and 7, respectively. However, the

Anderson-Darling test has p-values smaller than 0.01 for Labor cost and Inventory quality,

denoting a not-normal distribution. Indeed, the histogram shows these variables with

These tests are carried out for all 40 indicators. For skewness and kurtosis measurement,

we do not identify values higher than the limit determined. For the Anderson-Darling test,

the results demonstrate 14 indicators with not-normal distributions from the 40 variables

analyzed (see Appendix G for all test results). Nevertheless, this result does not impede

practical situations the warehouses do not always provide normal data, we consider that

these data characteristics are similar to reality to perform the aggregation analysis.

The correlation measurement results are evaluated in parallel with the theoretical model

of relationships (Jacobian matrix) to dene the indicators that should be discarded of the

analysis and the ones that will make part of the integrated model.

The correlation matrix, calculated using the standardized data, is presented in Table

6.5 and the numbers inside it are the correlation coecients, named Person's r (or just r ).

All highlighted cells present a signicant correlation, with p-value < 0.01. The blue cells

present the absolute value of the medium correlations, established between 0.4 up to 0.59;

and the pink cells show the absolute value for high correlations, determined from 0.6 up to

1.

We can verify that some indicators in Table 6.5 have weak or a few medium correlations.

For example, EqDp , Invq and Maintc do not have correlations higher than 0.4 (|r| ≥

0.4). These indicators might have problems to be incorporated in the results of PCA,

since the components are arranged based on the correlations between variables. This

result is evaluated in Chapter 7 with the complete group of informations coming from the

6.4. Statistical Tools Application 113

114

Table 6.5: Data correlation matrix.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40

1 CSc 1

2 CustSatq -0,1 1

3 Delp -0,6 0 1

4 Delt 0,65 -0 -0,98 1

5 DSt 0,02 -0,1 -0,16 0,13 1

6 EqDp -0,1 0,1 0,07 -0,1 -0,2 1

7 Invc 0,1 -0,2 -0,04 0,03 0,16 -0 1

8 Invq -0,1 -0,1 0,07 -0,1 -0,1 -0 0,09 1

9 InvUtp 0,11 -0,2 -0,02 0,02 0,16 -0 0,97 0,1 1

10 Labc 0,55 -0,2 -0,52 0,52 0,1 -0 0,14 -0 0,12 1

11 Labp -0,96 0,1 0,67 -0,7 -0,1 0,1 -0,13 0,1 -0,1 -0,7 1

12 Maintc -0 -0,1 0,11 -0,1 -0 -0 0,03 -0 0,04 0,21 0,07 1

13 OrdFq 0,04 -0,2 0 0,01 0 -0 0,11 0,1 0,1 0,06 -0 -0 1

14 OrdLTt 0,65 -0 -0,98 1,0 0,13 -0 0,03 -0 0,02 0,52 -0,7 -0,1 0 1

15 OrdProcc 0,60 0 -0,97 0,99 0,14 -0 0,04 -0 0,02 0,43 -0,6 -0,1 0 0,99 1

16 OTDelq -0 0,6 -0,15 0,12 0,05 -0 -0,05 0 -0 -0 -0 -0 -0,2 0,12 0,12 1

17 OTShipq 0,07 -0,2 -0,03 0,04 -0 -0 0,03 0 0,03 0,06 -0 -0 0,9 0,04 0,05 -0,2 1

18 PerfOrdq -0 0,7 -0,03 0,03 -0 0,1 -0,03 -0 -0 -0 -0 -0 -0,1 0,03 0,02 0,9 -0,2 1

19 Pickp -0,6 0 1 -0,97 -0,2 0,1 -0,05 0,1 -0 -0,5 0,66 0,11 0 -0,97 -0,97 -0,1 -0 -0 1

20 Pickq -0,1 0,1 0,07 -0,1 -0 0,1 -0,18 -0 -0,1 -0,1 0,11 -0,1 -0,2 -0,1 -0,1 0,1 -0,1 0,1 0,04 1

21 Pickt 0,63 -0 -0,97 1,00 0,13 -0 0,03 -0 0,02 0,51 -0,7 -0,1 0 1,0 0,99 0,1 0 0 -0,98 -0,1 1

22 Putt 0,38 -0,1 -0,32 0,33 -0,2 -0 -0,12 -0 -0,1 0,74 -0,5 0,12 0 0,33 0,24 -0,1 0,1 -0,1 -0,3 -0 0,31 1

23 Recp -0,39 0,1 0,34 -0,3 0,15 0,2 0,11 0,1 0,14 -0,76 0,51 -0,1 0 -0,3 -0,3 0,1 -0,1 0,1 0,33 0,01 -0,3 -1 1

24 Recq 0,08 0,1 -0,21 0,19 0,1 -0 0,01 0,2 0,04 0 -0,1 -0,1 -0,1 0,19 0,22 0,1 -0,1 0,1 -0,2 0,09 0,21 -0,1 0,04 1

25 Rect -0 -0,1 -0,12 0,1 1,00 -0 0,17 -0 0,17 0,02 -0 -0 0 0,1 0,11 0,1 -0 -0 -0,1 -0 0,1 -0,3 0,24 0,1 1

26 Repp -0,95 0,1 0,69 -0,7 -0,1 0,1 -0,14 0,1 -0,1 -0,69 0,98 0,07 -0 -0,7 -0,7 0 -0,1 0 0,68 0,08 -0,7 -0,5 0,5 -0,1 -0 1

27 Repq -0,1 -0 0,06 -0 0,03 -0 0,18 0,1 0,19 -0,1 0,11 0,07 0,1 -0 -0 0 0,1 -0 0,06 -0,1 -0 -0,1 0,13 0,1 0,04 0,04 1

28 Rept 0,96 -0,1 -0,7 0,73 0,06 -0 0,13 -0 0,12 0,69 -0,98 -0,1 0 0,73 0,68 -0 0,1 -0 -0,7 -0,1 0,72 0,47 -0,5 0,1 0,01 -0,99 -0 1

29 Scrapq 0,08 -0,3 0,03 -0 -0 0 -0,02 0,1 -0 0,12 -0,1 0,09 -0,2 -0 -0 -0,4 -0,3 -0,4 0,04 -0,3 -0 0,12 -0,1 -0,4 -0 -0 -0,41 0,03 1

30 Shipp -0,6 0 1,0 -0,98 -0,2 0,1 -0,05 0,1 -0 -0,5 0,67 0,11 -0 -0,98 -0,97 -0,1 -0,1 -0 1,0 0,07 -0,97 -0,3 0,34 -0,2 -0,1 0,69 0,06 -0,7 0,04 1

31 Shipq 0,05 -0,1 0,03 -0 -0 -0 0,07 -0 0,08 0,06 -0 0 0,8 -0 -0 -0,1 0,8 -0 0,05 -0,2 -0 0,02 -0 -0,2 -0 -0 0,03 0,02 -0,3 0,01 1

32 Shipt 0,65 -0 -0,98 1,00 0,13 -0 0,03 -0 0,02 0,52 -0,7 -0,1 0 1,00 0,99 0,1 0,1 0 -0,97 -0,1 1 0,33 -0,3 0,2 0,09 -0,7 -0 0,73 -0 -0,98 -0 1

33 StockOutq 0,06 -0,1 -0,04 0,03 0,08 -0 0,4 0 0,19 0,11 -0,1 -0,1 0,1 0,03 0,02 -0,1 0,1 -0 -0 -0,5 0,01 0,04 -0,1 -0,1 0,07 -0,1 0,09 0,08 0,05 -0 0,07 0,03 1

34 Stop -0,4 0,1 0,32 -0,3 0,16 0,2 0,14 0,1 0,16 -0,7 0,48 -0,1 -0 -0,3 -0,2 0,1 -0,1 0,1 0,31 0,01 -0,3 -0,99 0,99 0,1 0,25 0,48 0,14 -0,47 -0,1 0,32 -0 -0,3 -0 1

35 Stoq -0 0,2 0,06 -0,1 -0,2 -0 -0,15 0,2 -0,1 -0,2 0,04 -0,2 0,2 -0,1 -0 0,1 0,1 0 0,06 0,01 -0 -0,1 0,18 0,1 -0,1 0,05 0,01 -0,04 -0,46 0,06 0,2 -0 -0,1 0,14 1

36 Thp -0,96 0,1 0,67 -0,7 -0,1 0,1 -0,13 0,1 -0,1 -0,69 1 0,07 -0 -0,7 -0,6 -0 -0 -0 0,66 0,11 -0,7 -0,5 0,51 -0,1 -0 0,98 0,11 -0,98 -0,1 0,67 -0 -0,7 -0,1 0,48 0,04 1

6. Modeling

37 TOp -0,4 0,1 0,17 -0,2 -0,1 0 -0,88 -0 -0,91 -0,1 0,38 0,08 -0,1 -0,2 -0,2 0 -0,1 0 0,17 0,13 -0,2 0,14 -0,1 -0,1 -0,1 0,39 -0,2 -0,38 0,04 0,17 -0,1 -0,2 -0,2 -0,2 0,07 0,38 1

38 Trc 0,59 0 -0,96 0,98 0,14 -0 0,01 -0 -0 0,41 -0,6 -0,1 -0 0,98 0,98 0,1 0 0 -0,96 -0,1 0,98 0,24 -0,3 0,2 0,11 -0,6 -0 0,66 -0 -0,97 -0 0,98 0 -0,2 -0 -0,6 -0,2 1

39 TrUtp -0,5 0,2 0,4 -0,4 -0,1 0,2 -0,08 0,2 -0,1 -0,96 0,64 -0,2 -0 -0,4 -0,3 0,1 -0 0,1 0,37 0,07 -0,4 -0,8 0,77 0,1 0,02 0,62 0,09 -0,62 -0,1 0,4 -0,1 -0,4 -0,1 0,75 0,18 0,64 0,04 -0,3 1

40 WarUtp -0,1 -0,1 0,01 -0,1 0,12 -0 0,6 0,1 0,61 0,14 0,01 0,01 0,1 -0,1 -0,1 -0,1 0 -0,2 0,01 -0,1 -0 0,06 -0,1 0,1 0,11 0,03 0,04 -0,04 0,04 0,01 0,12 -0,1 0,2 -0,1 -0,1 0,01 -0,5 -0,1 -0 1

6.4. Statistical Tools Application 115

This section performs the rst PCA tests considering all variables in the model, the ones

The free software R is used to attain the results. There are two mathematical methods

is done with the eigenvalues of the correlation or covariance matrix, using the divisor N

(N = number of variables) for that. The prcomp formula, on the other hand, calculates a

singular value decomposition (centered and possibly scaled) of the data matrix, using the

of calculation for numerical accuracy. Thus, we use this one to perform our analysis.

Two main analysis are made in this rst phase: an analysis of indicators separated by

their dimensions of cost, quality, time, productivity (Section 6.4.3.1) and a global PCA

with all 40 indicators (Section 6.4.3.2). The objective is to verify the indicator's behavior

in aggregation situations, providing more elements to dene the nal group which will

As justied above (Section 6.4.1), data is standardized before their utilization in PCA

Initially, the PCA results for indicators separated by dimensions are shown in Figure 6.6

for quality, Figure 6.7 for productivity, Figure 6.8 for time, Figure 6.9 for cost. All gures

are divided in three parts (as well as Figure 6.10, showing the PCA result for all 40

cumulative proportion for the main components (in the bottom of the gure); a table of

indicators versus components (located in the up-left-side of the gures); the scree plot in

the right side of the gures. Each of these three parts is explained as follows.

The tables on the bottom of the gures have three dierent informations to analyze.

Initially, the standard deviation of each principal component higher than one is used as

Figure 6.6 there are 5 components (from PC1 up to PC5) with standard deviation higher

than one, indicating that these ve components should be considered in the representation

the contribution of each component to explain the data variance, whereas the cumulative

proportion (third line) presents the sum of all component variances. For Figure 6.6, the

cumulative proportion is 76,9%, signifying that the rst ve components explain 76,9% of

The indicator versus component tables demonstrate in the cells the loadings aij , giving

the weight of each indicator in the respective component. The highlighted cells are the ones

with |loading| ≥ 0, 3, denoting the indicators considered in each component. For example,

Figure 6.9 shows PC1 and PC2 (both with standard deviation higher than one) formed

by the following indicators: CSc , Labc , OrdProcc , Trc for PC1 and Invc , Labc and

Maintc for PC2. The linear combinations of indicators obtained from this table are shown

in Equation 6.1 and Equation 6.2. The signs of the loadings are arbitrary, and, according

to R documentation, they may dier between dierent PCA programs or even between

116 6. Modeling

dierent builds of R.

Finally, the scree plot shows the variance of the data (y axis, measured by the square

2 The principal

components are presented in decreasing order of importance with the objective of helping

analysts to easily visualize the sharp drop in the plot, which is also used as a signal that

One may expect from the PCA performed that each indicator dimension will be rep-

resented by one component (total of 4 components for all indicators), since indicators of

the same dimension could be more related among them than indicators of dierent dimen-

sions. Nevertheless, the results obtained do not conrm this hypothesis. The number of

components to include in the model (using the criterion of standard deviation higher than

one to retain components) are two for time and cost indicators, whereas for productivity

and quality are three and ve, respectively. It means that, if we would like to represent

all 40 indicators using these results, the number of components utilized will be 12 (2 of

expected. Since PCA has the objective to represent variables in a small number of prin-

cipal components, we can infer that 12 components are not a good result. Moreover, the

Looking at indicator versus component tables in Figures 6.6, 6.7, 6.8, 6.9, specically

see that a great quantity of indicators are allocated in more than one component, what

is not desirable for PCA results. The worst results can be seen for quality and produc-

tivity dimensions, with more than half of indicators allocated in at least two components.

Therefore, we conclude that indicators are not related just by their dimensions.

6.4. Statistical Tools Application

QUALITY INDICATORS

PC1 PC2 PC3 PC4 PC5 PC6 PC7 PC8 PC9 PC10

CustSatq -0,32 0,32 0,20 0,06 -0,17 0,15 -0,10 -0,18 -0,80 0,05

Invq 0,06 0,02 -0,20 -0,46 -0,52 -0,51 -0,31 0,26 -0,13 0,06

OrdFq 0,45 0,29 0,04 0,11 -0,05 -0,15 0,07 -0,03 -0,08 -0,31

OTDelq -0,34 0,35 0,26 0,04 -0,06 -0,28 -0,08 0,08 0,39 -0,10

OTShipq 0,45 0,29 0,01 0,15 0,03 -0,14 0,05 0,04 -0,13 -0,46

PerfOrdq -0,34 0,35 0,33 0,09 0,01 -0,20 0,02 0,03 0,21 -0,04

Pickq -0,18 0,07 -0,51 0,41 0,15 -0,04 -0,12 0,59 -0,14 0,09

Recq -0,16 0,15 -0,35 -0,34 -0,01 -0,17 0,76 -0,17 -0,06 0,06

Repq 0,05 0,14 -0,10 -0,47 0,61 -0,08 -0,46 -0,22 -0,05 0,05

ScrapRate 0,05 -0,49 0,23 0,14 -0,29 -0,21 -0,08 -0,21 -0,05 0,03

Shipq 0,41 0,33 0,10 0,17 -0,05 -0,07 0,04 -0,05 0,07 0,81

StockOutq 0,13 -0,06 0,49 -0,40 0,09 0,26 0,21 0,64 -0,11 0,03

Stoq 0,03 0,29 -0,23 -0,18 -0,45 0,63 -0,19 -0,08 0,29 -0,07

PC1 PC2 PC3 PC4 PC5 PC6 PC7 PC8 PC9 PC10 PC11

Standard

1,77 1,6708 1,2488 1,2011 1,033 0,915 0,887 0,7366 0,613 0,387 0,362

deviation

Proportion of

0,241 0,2147 0,12 0,111 0,082 0,064 0,06 0,0417 0,029 0,012 0,01

Variance

Cumulative

0,241 0,4557 0,5757 0,6867 0,769 0,833 0,894 0,9354 0,964 0,976 0,986

Proportion

117

118

PRODUCTIVITY INDICATORS

Delp 0,34 0,02 0,33 -0,28 -0,14 -0,07 0,05 0,15

EqDp 0,07 0,00 -0,23 -0,59 0,76 0,05 -0,11 0,01

InvUtp -0,05 -0,58 0,20 0,04 0,04 0,43 -0,16 0,42

Labp 0,37 0,06 0,01 0,30 0,18 0,24 -0,08 0,18

Pickp 0,33 0,02 0,34 -0,29 -0,15 -0,09 0,00 -0,23

Recp 0,27 -0,27 -0,40 -0,03 -0,21 -0,22 -0,29 -0,06

Repp 0,37 0,06 0,03 0,29 0,18 0,17 -0,12 -0,63

Shipp 0,34 0,02 0,33 -0,28 -0,14 -0,08 0,04 0,15

Stop 0,26 -0,29 -0,40 -0,03 -0,20 -0,27 -0,31 0,09

Thp 0,37 0,06 0,01 0,30 0,18 0,24 -0,08 0,18

TOp 0,12 0,58 -0,08 0,17 0,08 -0,30 -0,17 0,49

TrUtp 0,30 -0,13 -0,36 0,02 0,00 0,00 0,85 0,12

WarUtp -0,03 -0,38 0,35 0,33 0,42 -0,66 0,09 0,00

Standard

deviation 2,4838 1,5742 1,3505 0,9772 0,96 0,582 0,513 0,162

Proportion of

Variance 0,4745 0,1906 0,1403 0,0735 0,071 0,026 0,02 0,002

Cumulative

Proportion 0,4745 0,6652 0,8055 0,8789 0,95 0,976 0,996 0,998

6. Modeling

Figure 6.7: PCA results for productivity indicators.

6.4. Statistical Tools Application

TIME INDICATORS

Delt -0,45 0,00 0,16 0,14 0,30 -0,40 0,02 0,71

DSt -0,07 0,66 -0,25 0,04 -0,01 -0,04 -0,70 0,00

OrdLTt -0,45 0,00 0,16 0,14 0,30 -0,40 0,02 -0,71

Pickt -0,45 0,00 0,17 0,14 -0,86 -0,01 0,01 0,00

Putt -0,19 -0,30 -0,86 0,37 -0,01 0,01 0,07 0,00

Rect -0,05 0,68 -0,16 0,01 0,00 0,04 0,71 0,00

Rept -0,37 -0,08 -0,27 -0,88 -0,01 0,00 0,00 0,00

Shipt -0,45 0,00 0,16 0,14 0,28 0,82 -0,05 0,00

Standard 2,1833 1,4445 0,8962 0,5803 0,079 0,025 0,01 0,00

deviation

Proportion of

0,5958 0,2608 0,1004 0,0421 8E-04 8E-05 1E-05 0,00

Variance

Cumulative

0,5958 0,8567 0,957 0,9991 1 1 1 1,00

Proportion

119

120

COST INDICATORS

CSc -0,48 0,09 -0,01 0,42 -0,76 0,01

Invc -0,07 0,42 -0,88 -0,22 -0,01 -0,02

Labc -0,40 0,44 0,10 0,53 0,59 -0,03

Maintc 0,03 0,74 0,46 -0,45 -0,19 0,01

OrdProcc -0,55 -0,20 0,02 -0,37 0,13 0,71

Trc -0,55 -0,21 0,05 -0,39 0,10 -0,70

Standard

1,6802 1,0965 0,9837 0,7768 0,621 0,134

deviation

Proportion of

0,4705 0,2004 0,1613 0,1006 0,064 0,003

Variance

Cumulative

0,4705 0,6709 0,8322 0,9328 0,997 1

Proportion

6. Modeling

Figure 6.9: PCA results for cost indicators.

6.4. Statistical Tools Application 121

Another PCA is performed with all 40 indicators together, and the results are shown in

Figure 6.10. The informations presented in Figure 6.10 are the same as described pre-

viously for each dimension. The dierence is in the indicator versus component table,

which demonstrates only the columns of principal components (PC) with standard devia-

tion higher than one (the criterion used to choose the number of components to retain).

Moreover, the minimum loading value is reduced to 0.2 (|loading| ≥ 0.2). This limit is

empirically chosen based on the loading results for the rst component.

Dening the number of components to retain by the scree plot (in the right side of Figure

6.10), one could choose them as the rst two; PC1 and PC2. Indeed, these components

are which better contain/explain variable's variance, and the sharp drop is in that point in

the plot. From the standard deviation perspective, there are 10 components with standard

deviation higher than one, proposing the use of all of them to represent the indicators.

Comparing the results with respect to two or ten components we can see that with 2

components, 19 indicators are excluded from the analysis and with 10 components none of

them is excluded. Regarding the number of indicators designated for several components,

are allocated in more than one PC. Moreover, two components explain 44% of data variation

whereas ten components represent 86%. This situation establishes a trade-o between both

options.

As the analysis carried out on this thesis objectives to aggregate the greater number of

indicators as possible, we consider initially the 10 principal components in the model. The

main reason for this choice is that this result can be improved in Section 7.2 to get the

nal integrated model. However, in situations where there is a doubt about the number

and are in accordance to the warehouse reality. A framework to demonstrate the results

presented in indicator versus component table, of Figure 6.10, is built in Figure 6.11.

The names inside the blue rectangles are chosen according to the most relevant quantity

of indicator activities that the component encompasses. For example, C1 (which is derived

from PC1 column of Figure 6.10) is named Outbound Performance because the majority

of indicators making part of this component are related to replenishment, picking, shipping

and delivery activities. Also, C3 (representing the PC3 column of Figure 6.10) is dened

as Inventory Utilization because indicators related to stocks and space utilization are

comprised in the component. The exception is C2, named Mixed Performance because

half of the indicators are linked to the delivery and the other half to inbound activities.

This component in particular does not present a good result, since there is no physical

relation among these outbound and inbound indicators (it is possible to see it in the

Jacobian and correlation matrix). It probably happens because there are indicators with

just very low correlations, and their data confuse the PCA tool during the establishment

of indicator relationships. In Section 7.2 this result is analyzed again and these indicators

may probably be discarded of the analysis to improve the nal PCA result.

From Figure 6.11, we note a tendency in indicators' aggregation: the majority of indi-

cators are related in components according to their measurement domain. It means that

indicators are usually grouped with others from dierent dimensions but all metrics are

122 6. Modeling

PC1 PC2 PC3 PC4 PC5 PC6 PC7 PC8 PC9 PC10

CSc -0,22 0,06 -0,05 0,01 -0,08 0,14 -0,36 0,01 -0,05 0,07

CustSatq 0,02 -0,21 0,18 -0,02 -0,36 -0,03 0,02 0,18 0,01 -0,15

Delp 0,25 0,14 -0,03 0,01 -0,08 0,01 -0,12 -0,02 -0,04 0,02

Delt -0,26 -0,13 0,03 -0,02 0,09 0,01 0,10 0,03 0,03 0,03

DSt -0,03 -0,12 -0,15 0,13 0,13 -0,55 -0,16 -0,07 -0,07 -0,12

EqDp 0,04 -0,08 0,07 -0,02 0,04 0,28 -0,04 0,15 0,32 0,03

Invc -0,03 -0,01 -0,44 0,15 -0,15 0,07 0,15 0,04 0,08 -0,02

Invq 0,03 -0,05 -0,07 -0,06 0,05 0,26 0,19 -0,16 -0,34 -0,17

InvUtp -0,02 -0,03 -0,43 0,15 -0,17 0,07 0,12 -0,06 0,18 0,02

Labc -0,198 0,24 -0,01 0,05 -0,12 -0,16 0,07 -0,02 -0,02 0,02

Labp 0,24 -0,08 0,05 -0,03 0,11 -0,08 0,29 0,01 0,05 0,01

Maintc 0,01 0,13 0,00 0,07 -0,03 -0,15 0,14 0,07 0,03 0,58

OrdFq -0,01 0,06 -0,19 -0,49 0,05 -0,09 -0,01 0,06 0,10 -0,03

OrdLTt -0,26 -0,13 0,03 -0,02 0,09 0,01 0,10 0,03 0,03 0,03

OrdProcc -0,24 -0,17 0,02 -0,03 0,11 0,03 0,11 0,02 0,03 0,02

OTDelq -0,02 -0,22 0,13 0,02 -0,42 -0,15 0,08 0,20 0,03 -0,04

OTShipq -0,02 0,07 -0,16 -0,50 0,05 -0,10 -0,02 0,03 0,11 0,04

PerfOrdq 0,00 -0,20 0,12 0,02 -0,46 -0,12 -0,02 0,27 0,07 0,01

Pickp 0,25 0,14 -0,03 0,01 -0,09 0,01 -0,13 0,01 -0,04 0,02

Pickq 0,03 -0,05 0,14 0,04 -0,10 -0,04 -0,05 -0,50 0,45 0,01

Pickt -0,25 -0,14 0,04 -0,02 0,10 0,01 0,11 0,01 0,04 0,02

Putt -0,14 0,34 0,13 -0,02 -0,11 -0,05 0,19 -0,05 -0,01 -0,10

Recp 0,15 -0,33 -0,13 0,00 0,11 0,06 -0,19 0,07 0,02 0,10

Recq -0,04 -0,18 0,01 0,02 -0,11 0,01 0,08 -0,41 -0,25 -0,08

Rect -0,02 -0,15 -0,16 0,13 0,14 -0,53 -0,17 -0,06 -0,07 -0,11

Repp 0,24 -0,07 0,05 -0,02 0,09 -0,08 0,27 0,05 0,06 -0,04

Repq 0,02 -0,07 -0,13 -0,07 -0,08 0,00 0,16 -0,13 -0,33 0,58

Rept -0,24 0,07 -0,05 0,01 -0,08 0,09 -0,26 -0,04 -0,06 0,06

Scrapq -0,01 0,18 0,00 0,28 0,33 0,11 -0,08 0,29 0,10 -0,14

Shipp 0,25 0,13 -0,03 0,03 -0,09 0,02 -0,12 -0,02 -0,04 0,02

Shipq 0,00 0,07 -0,17 -0,48 -0,03 -0,12 -0,05 0,11 0,14 -0,04

Shipt -0,26 -0,13 0,03 -0,04 0,09 0,01 0,10 0,03 0,04 0,03

StockOutq -0,02 0,06 -0,19 0,03 0,00 0,00 0,13 0,42 -0,38 -0,14

Stop 0,14 -0,33 -0,14 0,02 0,11 0,06 -0,19 0,05 0,01 0,11

Stoq 0,03 -0,10 0,04 -0,27 -0,12 0,13 -0,10 -0,19 -0,29 -0,26

Thp 0,24 -0,08 0,05 -0,03 0,11 -0,08 0,29 0,01 0,05 0,01

TOp 0,07 0,06 0,41 -0,11 0,17 -0,19 0,08 0,02 -0,12 -0,02

Trc -0,24 -0,18 0,04 -0,03 0,11 0,02 0,12 0,03 0,03 0,02

TrUtp 0,17 -0,29 -0,01 -0,05 0,13 0,17 -0,03 0,01 0,01 -0,03

WarUtp 0,00 0,06 -0,30 0,09 -0,08 0,00 0,32 -0,13 0,15 -0,30

PC1 PC2 PC3 PC4 PC5 PC6 PC7 PC8 PC9 PC10 PC11 PC12

Standard

3,6571 2,0645 1,9917 1,7033 1,563 1,431 1,2734 1,24 1,15 1,063 0,99 0,882

deviation

Proportion

0,3344 0,1066 0,0992 0,0725 0,0611 0,051 0,0405 0,0387 0,0331 0,0282 0,0245 0,019

of Variance

Cumulative

0,3344 0,4409 0,5401 0,6126 0,6737 0,725 0,7654 0,8042 0,8372 0,8655 0,89 0,909

Proportion

6.5. Conclusions 123

Shipp Rept

Shipt C7-Outbound Productivity

Pickp Labp

InvUtp Invc

Pickt Trc WarUtp

CSc

TOp

Labc TrUtp C3 – Inventory Utilization

Stop OTDelq

OrdFq OTShipq

C5- Order Quality Scrapq

C2- Mixed Performance

C4-Ship Quality Shipq

Putt

CustSatq

Recp

PerfOrdq C8-Inventory Avaliability

Stoq

DSt

C6-Inbound Time

Recq Pickq StockOutq

Rect EqDp C10- Stock Quality

Productivity indicators

Invq C9-Product movement quality Repq Maintc

Cost indicators

Quality indicators

Time indicators Cut off level of 0,2 in absolute sense

from the same warehouse area. C1, for example, are formed of productivity, time and cost

indicators, and all of them are related to outbound activities. There are some exceptions

among the quality indicators: C4, C5 and C8 are components containing just quality indi-

cators. This is particularly interesting since these indicators also share data with indicators

Comparing the results of PCA performed for each dimension and for all indicators at

the same time, we conclude that the second analysis has provided better outcomes if the

components are compared in a practical sense. It means that the indicators aggregated

in components without dimension distinction seem to be more consistent with the reality.

Thus, the PCA result for all 40 indicators is used in the next chapter as the basis to the

integrated model development. As indicators with low correlations are also included in

the framework presented, the next chapter analyzes the right indicators that should be

6.5 Conclusions

In this chapter, we have created a scenario for the standard warehouse to generate the

data used to calculate indicators. This scenario represents the warehouse shop-oor with

its ow of products throughout the processes. An Excelr spreadsheet is elaborated with

data following normal and random distributions, which demonstrate the eect of chained

processes. This initial dataset is used to calculate performance indicators, which are em-

124 6. Modeling

A data sample of one month and the complete analytical model are coupled with

CADES

r software to calculate the Jacobian matrix. The assessment of the Jacobian ma-

trix makes part of an exhaustive procedure developed to infer about indicator relationships,

which calculates the partial derivatives matrix of the complete analytical model, encom-

From the results attained, we can conclude that it is very hard to quantitatively de-

termine from the partial derivatives the intensity of the relationship between indicators.

The procedure described in this chapter is, therefore, used to qualitatively analyze their

Further, the whole dataset (100 months) of indicator measures are utilized to apply

statistical tools. The correlation matrix and the principal component analysis are the

main tools performed to determine indicator relationships quantitatively and how they

could be aggregated to estimate the integrated performance. The PCA does not provide

good results in the dimensions aggregated separately nor in the total group of indicators.

The problems are mainly related to inconsistencies in the indicators group (some indicators

of the same component have no relationship among them) and to the great quantity of

indicators designated in more than one component. One reason for these problems may

be the variables not correlated with others, which can lead to misunderstandings of the

statistical model. Thus, next chapter evaluates these variables proposing an improved

Chapter 7

Model Solving, Implementation and Update

Unknown

Contents

7.2 Analysis of Jacobian and Correlation matrix to improve PCA

results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

7.3 Integrated performance model proposition . . . . . . . . . . . . 133

7.4 Scale for the Integrated Indicator . . . . . . . . . . . . . . . . . . 136

7.4.1 The analytical model adjustment . . . . . . . . . . . . . . . . . . . . . . 136

7.4.2 Objective function denition . . . . . . . . . . . . . . . . . . . . . . . . . 138

7.4.3 The choice of the optimization algorithm . . . . . . . . . . . . . . . . . . 139

7.4.4 The setting of the optimization tool . . . . . . . . . . . . . . . . . . . . . 139

7.4.5 The integrated indicator scale . . . . . . . . . . . . . . . . . . . . . . . . 142

7.5 Integrated Model Implementation . . . . . . . . . . . . . . . . . . . 145

7.6 Model Update . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

7.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

Abstract

This chapter proposes the integrated model to evaluate warehouse performance. To

attain this objective, the results obtained from dierent sources are analyzed to de-

termine the best number of components to consider in the model. Moreover, a scale

is developed for the integrated model utilizing an optimization tool, which denes the

upper and lower limits of the scale from the maximization and minimization results.

The integrated model with the scale is tested in two dierent warehouse performance

situations verifying that the utilization of the integrated model can help managers to

better evaluate the warehouse as a whole.

126 7. Model Solving, Implementation and Update

7.1 Introduction

The last chapter has presented the application of some methods to analyze indicator rela-

tionships. The results obtained with the measurement of relationships using the Jacobian

matrix, correlation matrix and PCA method are analyzed to propose an integrated perfor-

mance model.

performance evaluations. In order to help the interpretation of the integrated model results,

a scale is also proposed using the analytical model as a basis to perform an optimization,

which denes the upper and lower limits of the integrated indicator.

Afterwards, the utilization of the nal model with the developed scale is detailed, along

PCA results

Chapter 6 presents indicator relationships measured by the Jacobian matrix, the Correla-

tion matrix and the Principal Component analysis. To attain the nal integrated model,

the Jacobian and Correlation matrix are used as decision support to improve the PCA

From the PCA performed for all 40 indicators, presented in Section 6.4.3.2, we have

veried that some indicators do not t well the model, probably because of their low

could be seen as a high number considering the 40 input variables. In cases like that,

the analyst should nd the best balance between simplicity (retaining as few as possible

components, which cause the exclusion of indicators) and completeness (explain most of

data variation).

In this thesis, the initial suggestion of which indicators should be discarded of the

model come from an analysis of the Jacobian and Correlation matrix. Initially, we list the

worst outcomes obtained in the Jacobian (using Table 6.4) and in the Correlation matrix

(using Table 6.5). For the Jacobian, the worst results are represented by indicators with

the lowest number of shared data and, for the Correlation matrix, the indicators with the

lowest correlation values are the worst results. Secondly, the two lists generated with the

worst results are compared to suggest which indicators should be discarded of the model.

Table 7.1 summarizes these results in three parts: on the top of the table are presented

the indicators with bad results in both analysis; in the middle of the table, indicators with

bad results in correlation are listed with their corresponding number of shared data (from

Table 6.4) described in the right column; on the bottom of the table is the opposite: the

indicators with few number of shared data (from Table 6.4) are listed with their correlation

presented as follows.

From this initial list of 15 indicators presented in Table 7.1, we can see 4 of them

with no correlations higher than 0.4 (r ≤ 0.4). As PCA does not t a good model with

variables having no signicant correlations, these indicators are the rst candidates to be

7.2. Analysis of Jacobian and Correlation matrix to improve PCA results 127

Table 7.1: The indicators with Correlation and Jacobian worst results.

EqDp |r |≤ 0.4 with all indicators Shares 1 data with 20 indicators

Maintc |r |≤ 0.4 with all indicators Shares 2 data with CSc

Recq |r |≤ 0.4 with all indicators Shares 2 data with 4 indicators

Repq |r |= 0.41 with Scrapq Shares 2 data with 3 indicators

Pickq |r |= 0.5 with StockOutq Shares 2 data with 4 indicators

Indicator Correlation worst results Jacobian results

Invq | r |≤ 0.4 with all indicators Shares 2 data with 14 indicators

Stoq | r |= 0.46 with Scrapq Shares 2 data with 7 indicators

| r |= 0.5 with Pickq and | r |= 0.4 Shares 1 data with 6 indicators, 2 data

StockOutq

with Invc with 3 indicators, 5 data with Invc

Shares 1 data with 5 indicators, 2 data

| r |= 0.4 with Recq , | r |= 0.41 with

Scrapq with 7 indicators, 4 data with Thp and

Repq and | r |= 0.46 with Stoq

Labp

Indicator Correlation results Jacobian worst results

Shipq | r |= 0.8 with OrdFq and OTShipq Shares 2 data with 7 indicators

| r |= 0.9 with OrdFq and | r |= 0.8

OTShipq Shares 2 data with 7 indicators

with Shipq

| r |= 0.9 with OTShipq and | r |= 0.8

OrdFq Shares 2 data with 7 indicators

with Shipq

| r |= 0.6 with OTDelq and | r |= 0.7

CustSatq Shares 2 data with 9 indicators

with PerfOrdq

| r |= 0.6 with CustSatq and | r |= 0.9

OTDelq Shares 2 data with 9 indicators

with PerfOrdq

| r |= 0.7 with CustSatq and | r |= 0.9

PerfOrdq Shares 2 data with 9 indicators

with OTDelq

discarded (EqDp , Maintc , Recq , Invq ). However, Invq shares data with a great quantity

of indicators, demanding a deeper analysis. To determine the sequence of exclusion for the

indicators, we use the decreasing order presented in Table 7.1 (i.e. the indicators with no

correlations higher than 0.4 (r ≤ 0.4) and few shared data are deleted rst).

Five aspects are considered in the analysis of PCA results: (i) the number of principal

components (PC) with σ>1 should be the fewest possible; (ii) the cumulative proportion

of data explained by the PC's should be as high as possible; (iii) the number of indicators

designated in more than one component should be as low as possible; (iv) the loading signs

should be in accordance with indicator's objectives; (v) the indicators grouped in each

component should have a physical explanation in a warehouse context. These ve criteria

come from the literature about PCA application. The rst three aspects are quantitative

and used throughout the analysis of indicator's exclusion. The last two are evaluated at the

moment that the exclusion of an indicator provides only few changes in the quantitative

aspects.

In the cases that the indicator exclusion does not improve PCA result, the indicator is

maintained in the model and the following one of the list is tested. Therefore, all indicators

Table 7.2 shows the outcomes for each PCA, detailing the three quantitative parameters

128 7. Model Solving, Implementation and Update

Eliminated dicators left

0 40

• 10 PC with σ > 1

• 17 indicators designated in more than one com-

ponent

• Cumulative proportion of 10 PC: 86,5%

1 EqDp , 38

Maintc • 10 PC with σ > 1

• 18 indicators designated in more than one com-

ponent

• Cumulative proportion of 10 PC: 89%

2 Recq 37

• 9 PC with σ > 1

• 14 indicators designated in more than one com-

ponent

• Cumulative proportion of 9 PC: 88,6%

3 Repq 36

• 9 PC with σ > 1

• 14 indicators designated in more than one com-

ponent

• Cumulative proportion of 9 PC: 90,5%

4 Stoq 35

• 8 PC with σ > 1

• 12 indicators designated in more than one com-

ponent

• Cumulative proportion of 8 PC: 89,3%

5 Pickq 34

• 7 PC with σ > 1

• 15 indicators designated in more than one com-

ponent

• Cumulative proportion of 7 PC: 87,5%

6 StockOutq 33

• 7 PC with σ > 1

• 15 indicators designated in more than one com-

ponent

• Cumulative proportion of 7 PC: 89,8%

7.2. Analysis of Jacobian and Correlation matrix to improve PCA results 129

Only the exclusions that have improved the PCA results are demonstrated in Table

7.2. The PCA result after the exclusion of the three worst indicators (EqDp , Maintc ,

Recq ) is improved with one PC less than step zero (see Table 7.2), fewer indicators in more

components than before and data explanation of 88,6%, in comparison of 86% in step zero.

At the end, the exclusion of the majority of indicators with low or medium correlations

in Table 7.1 (EqDp , Maintc , Recq , Repq , Stoq , Pickq and StockOutq ) improve the

PCA result, providing a higher cumulative proportion of data explanation and the decrease

number of PC's (from 10 to 7). Invq and Scrapq are the only exceptions, being kept in

Even if the PCA outcome for step 5 is not demonstrated, we highlight that StockOutq

is excluded from the nal group because it has not been designated for any PC, i.e. the

Analyzing the indicators not excluded from the analysis but listed in Table 7.1, we

might conclude that the informations provided by the correlation and the Jacobian are

complementary because some indicators with low correlations have a great quantity of

Finally, the group of indicators considered for the aggregated model are 33 from the

Figure 7.1 demonstrates that PC1 explains 40% of data variance (table in the bottom

of the gure) and incorporate almost half of the indicators (14 of 33 in total) (rst column

Initially, the indicators are considered in a component when the loadings are higher

than 0.2 (| loading|≥ 0.2). Nevertheless, this minimum loading value cause some problems

in component 2. The rst inconsistency is about the inclusion of TrUtp and OTDelq

indicators in the component two, where the majority of indicators are related to inbound

activities. The second problem is the sign of Rect , that should be negative instead of

positive. As the absolute loading values in component one are at least 0.22, we dene this

value as the new cut o level (|loading| ≥ 0.22). According to PennState (2015a), the

In the work of Lu and Yang (2010), they include in the model just loadings higher then

0,5; however, the authors have considered their criterion very conservative.

Switching the absolute cut o level value to 0.22, a new PCA result is obtained (see

Figure 7.2), with two main dierences from the previous result (Figure 7.1). Firstly, the

utilization of the delivery truck and all other indicators are related to inbound activities.

However, if this indicator is eliminated the global results of other components become

worst. Therefore, the indicator is maintained in the nal model. Secondly, changing the

cut o level reduces to 8 the number of indicators designated in more than one component,

The sign of the loadings in Figure 7.2 should be in accordance with the indicator

objectives. In the case of cost and time indicators, the sign must be negative, whereas for

productivity and quality ones, the sign must be positive to represent a better performance.

In the case of component equations (presented in the next section) sharing both types of

loadings, they should be interpreted considering that the greater the resulting value, the

130 7. Model Solving, Implementation and Update

CSc -0,22 -0,05 0,08 -0,05 0,07 -0,09 0,38

CustSatq 0,02 0,17 -0,22 -0,01 0,40 0,00 -0,04

Delp 0,25 -0,14 0,06 -0,04 0,07 0,01 0,12

Delt -0,26 0,13 -0,06 0,05 -0,07 -0,03 -0,09

DSt -0,03 0,18 0,12 -0,07 -0,13 0,61 0,05

Invc -0,03 0,10 0,43 -0,17 0,13 -0,07 -0,18

Invq 0,03 0,05 0,07 0,03 -0,08 -0,31 -0,13

InvUtp -0,02 0,11 0,44 -0,17 0,15 -0,08 -0,16

Labc -0,20 -0,24 0,05 -0,07 0,10 0,18 -0,10

Labp 0,24 0,08 -0,07 0,07 -0,09 0,02 -0,29

OrdFq -0,01 -0,05 0,22 0,51 0,04 0,04 0,01

OrdLTt -0,26 0,13 -0,06 0,05 -0,07 -0,03 -0,09

OrdProcc -0,24 0,17 -0,06 0,06 -0,10 -0,06 -0,10

OTDelq -0,02 0,21 -0,18 -0,03 0,46 0,09 -0,09

OTShipq -0,02 -0,07 0,18 0,53 0,03 0,03 0,04

PerfOrdq 0,00 0,18 -0,17 -0,04 0,51 0,08 0,02

Pickp 0,25 -0,14 0,06 -0,04 0,08 0,02 0,13

Pickt -0,25 0,14 -0,06 0,05 -0,08 -0,04 -0,10

Putt -0,15 -0,37 -0,07 -0,02 0,09 0,07 -0,22

Recp 0,15 0,36 0,07 0,03 -0,09 -0,07 0,22

Rect -0,02 0,21 0,12 -0,07 -0,14 0,59 0,07

Repp 0,24 0,07 -0,07 0,06 -0,08 0,03 -0,28

Rept -0,25 -0,07 0,07 -0,04 0,07 -0,04 0,28

ScrapRate -0,01 -0,14 0,02 -0,24 -0,34 -0,06 0,06

Shipp 0,25 -0,13 0,05 -0,05 0,07 0,01 0,12

Shipq 0,00 -0,05 0,20 0,50 0,12 0,07 0,03

Shipt -0,26 0,13 -0,05 0,06 -0,07 -0,04 -0,09

Stop 0,15 0,37 0,08 0,01 -0,09 -0,07 0,21

Thp 0,24 0,08 -0,07 0,07 -0,09 0,02 -0,29

TOp 0,07 -0,13 -0,41 0,15 -0,15 0,17 -0,06

Trc -0,24 0,18 -0,08 0,06 -0,09 -0,05 -0,11

TrUtp 0,17 0,29 -0,04 0,08 -0,11 -0,20 0,06

WarUtp 0,00 -0,01 0,33 -0,11 0,05 0,01 -0,41

Standard

3,65 2,01 1,94 1,65 1,54 1,37 1,25 0,94

deviation

Proportion

0,40 0,12 0,11 0,08 0,07 0,06 0,05 0,03

of Variance

Cumulative

0,40 0,53 0,64 0,72 0,79 0,85 0,90 0,92

Proportion

Figure 7.1: PCA result for the nal group of 33 indicators with |loadings| ≥ 0.2.

7.2. Analysis of Jacobian and Correlation matrix to improve PCA results 131

CSc -0,22 -0,05 0,08 -0,05 0,07 -0,09 0,38

CustSatq 0,02 0,17 -0,22 -0,01 0,40 0,00 -0,04

Delp 0,25 -0,14 0,06 -0,04 0,07 0,01 0,12

Delt -0,26 0,13 -0,06 0,05 -0,07 -0,03 -0,09

DSt -0,03 0,18 0,12 -0,07 -0,13 0,61 0,05

Invc -0,03 0,10 0,43 -0,17 0,13 -0,07 -0,18

Invq 0,03 0,05 0,07 0,03 -0,08 -0,31 -0,13

InvUtp -0,02 0,11 0,44 -0,17 0,15 -0,08 -0,16

Labc -0,20 -0,24 0,05 -0,07 0,10 0,18 -0,10

Labp 0,24 0,08 -0,07 0,07 -0,09 0,02 -0,29

OrdFq -0,01 -0,05 0,22 0,51 0,04 0,04 0,01

OrdLTt -0,26 0,13 -0,06 0,05 -0,07 -0,03 -0,09

OrdProcc -0,24 0,17 -0,06 0,06 -0,10 -0,06 -0,10

OTDelq -0,02 0,21 -0,18 -0,03 0,46 0,09 -0,09

OTShipq -0,02 -0,07 0,18 0,53 0,03 0,03 0,04

PerfOrdq 0,00 0,18 -0,17 -0,04 0,51 0,08 0,02

Pickp 0,25 -0,14 0,06 -0,04 0,08 0,02 0,13

Pickt -0,25 0,14 -0,06 0,05 -0,08 -0,04 -0,10

Putt -0,15 -0,37 -0,07 -0,02 0,09 0,07 -0,22

Recp 0,15 0,36 0,07 0,03 -0,09 -0,07 0,22

Rect -0,02 0,21 0,12 -0,07 -0,14 0,59 0,07

Repp 0,24 0,07 -0,07 0,06 -0,08 0,03 -0,28

Rept -0,25 -0,07 0,07 -0,04 0,07 -0,04 0,28

Scrapq -0,01 -0,14 0,02 -0,24 -0,34 -0,06 0,06

Shipp 0,25 -0,13 0,05 -0,05 0,07 0,01 0,12

Shipq 0,00 -0,05 0,20 0,50 0,12 0,07 0,03

Shipt -0,26 0,13 -0,05 0,06 -0,07 -0,04 -0,09

Stop 0,15 0,37 0,08 0,01 -0,09 -0,07 0,21

Thp 0,24 0,08 -0,07 0,07 -0,09 0,02 -0,29

TOp 0,07 -0,13 -0,41 0,15 -0,15 0,17 -0,06

Trc -0,24 0,18 -0,08 0,06 -0,09 -0,05 -0,11

TrUtp 0,17 0,29 -0,04 0,08 -0,11 -0,20 0,06

WarUtp 0,00 -0,01 0,33 -0,11 0,05 0,01 -0,41

Standard

3,65 2,01 1,94 1,65 1,54 1,37 1,25 0,94

deviation

Proportion

0,40 0,12 0,11 0,08 0,07 0,06 0,05 0,03

of Variance

Cumulative

0,40 0,53 0,64 0,72 0,79 0,85 0,90 0,92

Proportion

Figure 7.2: PCA result for the 33 indicators with |loadings| ≥ 0.22.

132 7. Model Solving, Implementation and Update

The 7 variables excluded

EqDp

from the first PCA result Pickq Repq Stoq

Rept Thp

Labp

Repp C3- Shipping Quality

OTDelq Scrapq

Pickp CSc

Pickt Shipt

Trc OrdLTt C4- Order Quality

Invq Rect

TrUtp Recp Labc

PerfOrdq CustSatq

Stop C2- Inbound Performance Putt

TOp WarUtp

C5- Inventory Utilization InvUtp

Cost indicators

Quality indicators

Cut off level of 0,22 in absolute sense Invc

Time indicators

Figure 7.3: The indicators eliminated from the analysis and a framework of the nal group

with 33 indicators.

Regarding the number of PC's to use in the aggregated model, the scree plot suggests

that 2 components is a good trade-o between variance explained and number of compo-

nents (the sharp drop point in the plot). However, we want to maintain the same number

of indicators in the model. Analyzing indicator versus PC table of Figure 7.2, we can

see that PC7 is just a repetition of indicators already designated in previous components.

Thus, the performance indicators will be aggregated in the rst six components (from PC1

up to PC6). Figure 7.3 summarizes the results demonstrating on the top of the gure

the indicators eliminated from the model and on the center the nal framework with six

Analyzing the loading signs, we can see that some of cost and time indicators do not

have negative signs as expected and the same happens for some quality and productivity

indicators. For the six components, the loadings of C1, C2, C4 and C5 have the right signs

and the ones from C3 and C6 present the opposite signs compared to indicator's objective.

R documentation arms that the signs are dened arbitrarily and if it is necessary to

change them, it should be made for all loadings of the component. Therefore, the signs of

indicators in components three and six will be inverted when the component equations are

The next section presents the nal integrated model for warehouse performance man-

agement.

7.3. Integrated performance model proposition 133

Section 4.4.1 presents a generic group of equations to describe the integrated performance

model. In this section, these equations are rewritten according to the result obtained in

Figure 7.2. Equation 7.1 up to Equation 7.6 demonstrate the six components chosen with

their loadings. We recall that the signs of C3 and C6 are modied as explained in the

previous section. The modied signs are highlighted with red color in the equations.

− 0, 24 × OrdP rocc + 0, 25 × P ickp − 0, 25 × P ickt + 0, 24 × Repp (7.1)

(7.3)

−0, 33 × W arU tp

It is important to highlight that indicator values entries in Equation 7.1 up to 7.6 must

Once the standardized indicator results are inserted in equations, it reduces their vari-

ance, making possible to verify which indicators most inuence the component result

through the loading values. For example, in Equation 7.5 the indicators OTDelq and

PerfOrdq have the highest loading values, demonstrating that they are more important

in C5 than CustSatq and Scrapq . However, not all components have this distinction

between indicators. For instance, in the rst component equation (C1, Equation 7.1) the

loading values are very similar for all indicators, resulting nearly in the same absolute

Equation 7.1 up to Equation 7.6 shows the model to measure the integrated performance

just one component to evaluate performance, probably the most important for company's

goals. In this case, the aggregation stops here and the manager loses a great quantity of

134 7. Model Solving, Implementation and Update

analyze the warehouse performance, it is necessary to develop a scale for each component,

allowing the manager to evaluate each group of indicators separately. However, it does not

seem a practical choice if the objective is to analyze the global warehouse performance.

The component results are very subjective and dicult to compare with other components,

As the main idea of this work is to dene a model which aggregates all indicators

to facilitate the global performance interpretation, we propose the sum of all principal

i=m

X

GP = n i × Ci (7.7)

i=1

In this dissertation, the weight of each component is considered equal, and each ni of

1

Equation 7.7 is dened by ( ) (m = 6 in our case). Nevertheless, the manager can adjust

m

each weight according to company's goals and strategy, dening some of them as more or

to Equation 7.7. Figure 7.4 demonstrates the framework with indicators aggregated in

components (left side of the gure) and the components composing the global performance,

7.3. Integrated performance model proposition

The integrated performance measurement model for warehouse management

1 – The final aggregated model for warehouse indicators 2 – The proposed global indicator

Rept Thp

Labp

Repp C1 - Outbound Performance

C3- Shipping Quality

OTDelq Scrapq

Pickp CSc

C3- Shipping Quality GP

Pickt Shipt

Trc OrdLTt C4- Order Quality

C4- Order Quality

Invq Rect

TrUtp Recp Labc

PerfOrdq CustSatq C5- Inventory Utilization

Stop C2- Inbound Performance Putt

TOp WarUtp

C5- Inventory Utilization InvUtp

Cost indicators

Quality indicators

Cut off level of 0,22 in absolute sense Invc

Time indicators

Figure 7.4: The integrated performance measurement model comprises: (1) the nal aggregated model and, (2) the global indicator.

135

136 7. Model Solving, Implementation and Update

The procedure performed in this section can be used for one component (if the manager

considers just one) as well as for the proposed global performance GP (Equation 7.7). In

2. Objective function;

3. Optimization algorithm;

The rst analytical model, used for Jacobian matrix assessment, needs to be adjusted to

perform the optimization. The adjustments signify mainly the inclusion of new equations

The last two kinds of equations are presented in the next sections.

Equations 7.1 up to 7.6 request standardized indicator values to calculate components. As

the data inputs of the analytical model are not standardized, it is necessary to include equa-

tions which shift indicator values to standardized ones. Thus, 33 equations like Equation

7.8 (i.e., one for each indicator) are added to the model. The mean and standard deviation

values inserted for each indicator are taken from their data generation. The complete list

is presented in Appendix I.

OrdFq − M ean_OrdFq

OrdFq _N ORM = (7.8)

σOrdFq

Some equations dening data dependencies are included in the optimization model to limit

the optimization search space so that the results fall within reasonable practical values.

This is done by constraining some additional variables. These equations have also been

The complete list of equations and the optimization model are demonstrated in Ap-

pendix H.

7.4. Scale for the Integrated Indicator 137

As an example, let us analyze Equation A.3. If it is not included in the model, the opti-

mization algorithm treats the variables Cor Unlo and Prob Unlo as independents. However,

Other equations limiting the optimization search space are dened by the prex Ctrl.

One example is Ctrl_0 (Equation 7.9) that determines the total eective working hours

made by the administrative employees. These can not be higher than the total number of

(7.9)

+ H Adminship + H Admindel + H Adminorders

Other examples are related to the warehouse product ow, impeding that one activity

processes more products than the previous one. For instance, Ctrl_1 (Equation 7.10)

denes that the number of pallets stored can not be higher than the total of pallets unloaded

in the whole month. Other constraints similar to Ctrl_1 are: Ctrl_2, Ctrl_3, Ctrl_4,

Ctrl_5 (Equations 7.11, 7.13, 7.14, 7.16, respectively). Some terms used in these equations

The replenishment is the activity of reallocating pallets from the bulk storage area to

the forward picking area. Due to its characteristics, there are two constraints related to

this activity ( Ctrl_2 Ctrl_2A). As the forward picking stock usually has a limited

and

space, the products are not replenished if they do not have orders to be fullled (Ctrl_2,

Equation 7.11). Similarly, the total number of pallets moved to the forward area can not

exceed the number of pallets stored plus the inventory remaining from the previous month

(

Ctrl_2 → > Pal Moved (7.11)

Prod pal

Remain inv

Ctrl_2A → Pal Sto + > Pal Moved (7.12)

Prod pal

Regarding the number of orders picked during a month, Ctrl_3 shows that it can not

be higher than the number of customer orders received (Cust Ord). In Equation 7.13, Line

Ord means the average number of lines per customer order, being used to put the number

of order lines picked (OrdLi Pick) in the same unit of customer orders.

OrdLi Pick

Ctrl_3 → Cust Ord > (7.13)

Line Ord

138 7. Model Solving, Implementation and Update

The Ctrl_4 and Ctrl_4A have the same meaning, just the units are dierent. In

Equation 7.14, the number of orders shipped can not overcome the total of orders picked

OrdLi Pick

Ctrl_4 → > Ord Ship (7.14)

Line Ord

As presented for the previous warehouse activities, Ctrl_5 represents the limitations

imposed by the activity ows. Equation 7.16 determines that the number of orders shipped

(Ord Ship) is higher or equal to the number of orders delivered (Ord Del).

Finally, Ctrl_6 denes that the number of orders delivered on time (Ord Del OT) is

always greater than the number of orders delivered on time, without damages and correct

documents (Ord OT, ND, CD), since this last one demands more order requirements than

After the denition of optimization limits by these equations, it is missing only the

The objective function is determined by the GP equation, Equation 7.7, which calculates a

weighted mean of all components dened in PCA. The maximization (Equation 7.18) and

the minimization (Equation 7.19) of GP achieve the best and worst possible performances,

respectively, which are considered the upper and lower limits of the scale. As dened

in Section 7.3, we assume that the weights are dened equal for all components in GP

equation.

It is important to note that these best and worst performances are only related to the

warehouse studied, and can not be generalized to other warehouses. The main reason is

that the optimization search space is established according to the warehouse conditions

1

max GP = ( ) × [C1 + C2 + C3 + C4 + C5 + C6 ] (7.18)

6

1

min GP = ( ) × [C1 + C2 + C3 + C4 + C5 + C6 ] (7.19)

6

After the analytical model and objective function determination, we dene, in the next

7.4. Scale for the Integrated Indicator 139

The analytical model that has been created has many outputs that must be constrained

in order to solve the problem. Thus, we are interested in algorithms that are able to deal

with several constraints. To that end, the fast and deterministic SQP algorithm (Sequential

The main reason for that choice is the possibility to manage tens, hundreds or even

model with the SQP requires the determination of the Jacobian matrix associated to the

model outputs. The CADES Component Optimizer r has the SQP algorithm built in the

The optimization model implemented in CADES comprehends: the analytical model, the

When the model is compiled, it generates an icar component containing the input and

output relationships and the associated Jacobian matrix. In order to use the SQP to solve

the problem, the inputs must be set with an initial value. Additionally, the inputs can also

be left free to vary in a range, dening the optimization search space. The outputs can

be left free to vary, have a xed value assigned to it or constrained in a range. Figure 7.5

One of the potential problems that may arise from the utilization of the SQP is that

the solution may depend on the starting values of the inputs (local minimum). Therefore,

it is a good practice to test several combinations of these initial values in order to increase

the possibility of nding a global optimum. Such investigation is made to dene the initial

values of the inputs that are used in the optimization study presented on this chapter.

Regarding the limits proposed for variables, they need to t the conditions of the

the possible upper and lower limits that the warehouse can attain to develop achievable

scale boundaries. In our case, the variable limits are established based on some predened

warehouse characteristics (e.g. warehouse capacity) and according to the limits presented

CADES requires the denition of a minimum and a maximum value for each constraint

output. Therefore, the inequality equations must be rewritten. For example, Equation

7.16 is modied for the following form to dene the minimum value:

The maximum limit set in CADES for Equation 7.20 is determined by the maximum

allowed value of the Ord Ship. All the constraints are dened in CADES using the same

140 7. Model Solving, Implementation and Update

Figure 7.5: The options provided by CADES Component Optimizer r for input and output

variables.

principle.

The variables are classied as inputs, intermediate outputs and outputs. The input

limits shown in Table 7.3 are separated by type of data and the unit of each variable

is presented in brackets. Some variables are considered xed in the optimization, as the

To establish the limits presented in Table 7.3 and Table 7.4, we consider that the stan-

dard warehouse has capacity to process up to 40.000 products per month (the mean value

dened in data generation is 28.000 with standard deviation of 2.000), and a maximum of

3.000 orders. Transforming the 40.000 products in number of pallets (each pallet has 40

products), we have 1.000 pallets as inbound capacity for unloading and storing activities.

For replenishment, the limit is of 2.000 pallets because we consider the sum of the stock

We note that the variables Prob OrdLi Pick, Prob Rep, Prob Sto, Prob Unlo (see

Table 7.4) have a limit smaller than the ones dened for shipping and delivery activities

(20 pallets for Prob Sto and Prob Unlo instead of 1.000 pallets; 40 orders for Prob OrdLi

Pick and Prob Rep instead of 3.000 orders). The reason for this limit is the absence of

quality indicators related to these activities in component equations; consequently, the

2% of the total capacity as the maximum quantity of problems each activity can have (as

7.4. Scale for the Integrated Indicator 141

INPUT LIMITS

Limits in Hours Picking, Shipping and Limits

Time data [unit]

Max Min Delivery data [unit] Max Min

β_del 1 0,3 Prod noAvail [orders] 3000 0

β_ord 1 0,3 No_OT_del [orders] 3000 0

β_pick 1 0,3 No_OT_ship [orders] 3000 0

β_rec 1 0,3 No Cust Complain [orders] 3000 0

β_rep 1 0,3 NoComplet Ord Ship [orders] 3000 0

β_ship 1 0,3 Other_Prob_pick [orders] 40 0

β_sto 1 0,3 Other_Prob_del [orders] 3000 0

Hadmindel [hour] 210 1 Other_Prob_ship [orders] 3000 0

HAdminpick [hour] 210 1 Cor OrdLi Pick [orders] 3000 0

HAdminrec [hour] 210 1 Cor OrdLi Ship [orders] 3000 0

HAdminrep [hour] 210 1 Cor Del [orders] 3000 0

HAdminship [hour] 210 1 scrap4 [orders] 40 0

HAdminsto [hour] 210 1 scrap5 [orders] 3000 0

scrap6 [orders] 3000 0

Limits

Replenishment data [unit]

Max Min Unloading and Storing data Limits

Cor Rep [pallet] 2000 0 [unit] Max Min

error data system 3 [pallet] 40 0 Cor Sto [pallet] 1000 0

scrap3 [pallet] 40 0 Cor Unlo [pallet] 1000 0

Other_Prob_rep [pallet] 40 0 scrap1 [pallet] 20 0

scrap2 [pallet] 20 0

Limits in $ Other_Prob_sto [pallet] 20 0

Cost data

Max Min Other_Prob_unlo [pallet] 20 0

Maintc R$ 50 000,0 R$ 1 000,0 error data system 1 [pallet] 20 0

Truck Maint C R$ 200 000,0 R$ 50,0 error data system 2 [pallet] 20 0

Limits

Fixed data [unit] Values Other data [unit]

Max Min

N [nb of components] 6 War WH [hour] 210 80

pal_truck [pallet] 25 Prod Ord [product] 30 10

Prod Cost [R$] R$ 100,00 war used area [m2] 4000 1000

$ oil [R$] R$ 2,20 nb_Travel [travels] 300 1

mean_Insp [h] 1 0,1

Cust Ord [orders] 3000 10

142 7. Model Solving, Implementation and Update

Limits Limits

Constraints [unit] Data [unit]

Max Min Max Min

CTRL_0 [hour] 210 0,1 aveinv [product] 80000 1

CTRL_1 [pallet] 1000 0 Prob Data [pallet] 80 0

CTRL_2 [pallet] 2000 0 Cust Complain [orders] 3000 0

CTRL_2A [pallet] 2000 0 ΔT(Insp) [hour] FREE

CTRL_3 [order] 3000 0 nb_trucks [trucks] FREE

CTRL_4 [order] 3000 0 Prod noAvail [products] 50000 0

CTRL_4A [product] 50000 0 Ord Del OT [orders] 3000 0

CTRL_5 [order] 3000 0 Ord OT, ND, CD [orders] 3000 0

CTRL_6 [order] 3000 0 Ord Ship OT [orders] 3000 0

PalProcInv [pallets] 4000 0

Component Prob OrdLi Pick [orders] 40 0

Limits

Equation Prob OrdLi Ship [orders] 3000 0

C1 Prob Del [orders] 3000 0

C2 Prod Proc [products] 40000 0

C3 Prob Rep [pallet] 40 0

C4

FREE Prob Sto [pallet] 20 0

C5 Prob Unlo [pallet] 20 0

C6 Remain_Inv [products] 40000 0

WarCapUsed 5000 500

Pal Sto [pallet] 1000 300

Pal Unlo [pallet] 1000 300

Pal Moved [pallet] 2000 500

OrdLi Pick [orders] 3000 700

Ord Ship [orders] 3000 700

Ord Del [orders] 3000 700

The nal output limits are presented in Table 7.5. The range of indicator values are

dened very large and cost indicators are left free. The cost indicators are not constrained

The results for the maximization and minimization are presented in next section.

The maximization and minimization results for the nal outputs are shown in Table 7.6.

The maximization and minimization results for the inputs and the intermediate outputs

We establish that the number of warehouse working hours War WH could vary between

80 and 210 hours per month (see Table 7.3). In the maximization, the War WH converges

in 210 hours, which is equivalent to 25 working days in a month (see the last table at the

bottom of Appendix J). In the 80 hours, 40.000 products are shipped (Prod Proc ) and

for 210 hours just 8.790 products. It means that if time is eciently used, the excess of

7.4. Scale for the Integrated Indicator 143

Time Indicators Limits Productivity Limits

[unit] Max Min Indicators Max Min

OrdLTt [h/order] 500 0,05 Thp 1500 0

DSt [h/pallet] 200 0,02 Labp 200 0,1

Delt [h/order] 200 0,02 Delp 200 0,01

Pickt [h/order] 200 0,02 Pickp 200 0,01

Putt [h/pallet] 200 0,02 Recp 200 0,01

Rect [h/pallet] 200 0,02 Repp 200 0,01

Rept [h/pallet] 200 0,02 Shipp 200 0,01

Shipt [h/order] 200 0,02 Stop 200 0,01

TOp 50 0

Limits in % TrUtp 100% 0%

Quality Indicators

Max Min InvUtp 105% 0%

CustSatq 100% 0% WarUtp 100% 0%

Invq 100% 0%

OrdFq 100% 0% Limits in $

Cost Indicators

OTDelq 100% 0% Max Min

OTShipq 100% 0% CSc R$ 1,00 0,00

PerfOrdq 100% 0% Invc

Shipq 100% 0% Labc

Scrapq 100% 0% OrdProcc

FREE

Trc

Global Limits

Performance Max Min

GP 150,0 -150,0

Time Indicators Results Productivity Results

[unit] Maximization Minimization Indicators Maximization Minimization

OrdLTt [h/order] 0,09 2,79 Thp 500 41,8

DSt [h/pallet] 0,05 1 Labp 55,5 4,65

Delt [h/order] 0,02 0,6 Delp 18,75 1,67

Pickt [h/order] 0,03 1,2 Pickp 9,37 0,83

Putt [h/pallet] 0,02 0,69 Recp 25 3,43

Rect [h/pallet] 0,03 0,32 Repp 12,5 2,38

Rept [h/pallet] 0,03 0,422 Shipp 12,5 1,1

Shipt [h/order] 0,03 0,9 Stop 25 3,43

TOp 2 0,88

Results TrUtp 100% 5,9%

Quality Indicators

Maximization Minimization InvUtp 50% 25%

CustSatq 100% 0% WarUtp 32% 86%

OrdFq 100% 100%

OTDelq 100% 0% Results

Cost Indicators

OTShipq 100% 0% Maximization Minimization

PerfOrdq 100% 0% CSc R$ 0,09 1,00

Invq 100% 99,2% Invc R$ 200 000,00 R$ 100 000,00

Shipq 100% 0% Labc R$ 5 987,20 R$ 15 718,50

Scrapq 0% 37,6% OrdProcc R$ 0,15 R$ 0,54

Trc R$ 0,70 R$ 292,80

Global Results

Performance Maximization Minimization

GP 15,35 -123,27

144 7. Model Solving, Implementation and Update

As expected, the maximization results for time indicators are low and for the productiv-

ity indicators are high (see Table 7.6). The capacity measures InvUtp and WarUtp have

low values, demonstrating that the warehouse can process more products due to its extra

capacity. The InvUtp values (see Table 7.6) show the maximization having higher results

than minimization. The reason for these results is the quantity of products processed in

almost 5 times less than in maximization, which reduce the number of products that pass

through the inventory (see Appendix J). Consequently, the same occur for Invc indica-

tor, since in minimization the average inventory is of 10.000 and in maximization 20.000

products.

The Invq and OrdFq indicators present values in the minimization near to the maxi-

mum (see Table 7.6). In OrdFq case, the optimizer prioritizes the reduction of indicators

with the highest loadings in component equations. Another point is an optimization model

restriction, which impedes an order to have more than one kind of problem. As the load-

ings of OrdFq is 0,51 and of OTShipq is 0,53 (Equation 7.4), the software prefers to put

all orders shipped late but complete. In the case of Invq , the reason is the established 2%

as the maximum number of problems for unloading, storing, replenishment and picking

variation range is of 138,65, with the maximum of 15,35 and the minimum of -123,3.

The reason for this expressive dierence between the positive and negative values comes

from the mean and standard deviation established for performance indicators. As these

xed values are used to standardize the indicators included in component equations, when

the indicator value in a month is lower than its mean, the result of the standardized

indicator is negative. For example, Equation 7.8 presents the standardization of OrdFq .

Considering that the average of OrdFq is 97% and, at this month, OrdFq value is 95%,

the standardized indicator has a negative sign because the performance is lower than the

average. Therefore, analyzing Table I.1 in Appendix I, it is possible to see that the majority

of quality indicators have means equal or higher than 99%. It means that there is few space

for performance improvements, what is reected by the low value of 15,35 as the upper

scale limit.

To support the scale interpretation, we transform the scale limits from -123,3 and 15,35

Using traditional scale transformation rules, Equation 7.21 is used to transform the

= → NS = + 50 (7.21)

138, 65 100 138, 65

To exemplify the use of Equation 7.21, let us verify the corresponding value in the

normal scale (NS) for the zero value in the optimized scale (OS). The zero value in OS

signies that all indicators are equal to their mean. Applying Equation 7.21, the result

for the normal scale is 88,93. We can infer from this result that globally, the warehouse

In the next section is explained how the integrated model and scale are implemented

7.5. Integrated Model Implementation 145

Optimized Normal

Scale - OS Scale - NS

+15,35 100

-53,975 50

-123,3 0

After nishing the model and scale development, we present in this section how to use the

The model parts used for periodic management are: the 33 indicator equations (pre-

sented in Chapter 5); the 6 component and GP equations (Equations 7.1 up to 7.7);

optimized scale with transformation to normal scale (Equation 7.21). These equations can

be included in a spreadsheet to facilitate data update. Every month's indicator values are

For example, Table 7.7 demonstrates the results for all 33 indicators in two dierent

months. The component and GP values for each month (in optimized scale - OS - and

The indicator values are established in order to evaluate warehouse performance in two

dierent situations. In month 1, we consider that inbound activities have some performance

problems, aecting their indicators of time, productivity and quality (they are lower than

the average). In this special example, the indicators related to replenishment activity are

also considered with problems. The outbound indicators, on the other hand, have very

good results, higher than the average. In month 2 the opposite situation is established:

inbound indicators have good performance whereas outbound indicators have bad results.

It is interesting to note that the global performance of the rst month is better than

the second one. This result could maybe support some manager's practices preferring to

an updated model. Usually, new situations in the market impact on enterprises (and also

on their warehouses) requesting a reevaluation of the initial model. Described in the next

146 7. Model Solving, Implementation and Update

months.

Indicator

Month 1 Month 2

CSc 0,3 0,3

CustSatq 100 92

Delp 6 3,5

Delt 0,15 0,35

DSt 0,95 0,4

Invc 250000 70000

Invq 95 100

InvUtp 75 55 Table 7.8: GP result for two dierent months.

Labc 10500 16000

Labp 15 13

OrdFq 100 98

C Month 1 Month 2

OrdLTt 1 1,8

C1 6,24 -4,02

OrdProcc 0,6 0,92

OTDelq 100 94 C2 -5,16 3,41

OTShipq 100 94 C3 -0,99 -3,73

PerfOrdq 100 90 C4 2,69 -11,48

Pickp 2,8 1,8 C5 2,99 -23,79

Pickt 0,3 1

C6 -19,50 5,03

Putt 0,2 0,08

Recp 5 10 GP (OS) -2,29 -5,76

Rect 0,8 0,4 GP (NS) 87,28 84,77

Repp 3 5

Rept 0,3 0,1

Scrapq 2 7

Shipp 5 1,5

Shipq 99,5 96

Shipt 0,3 0,45

Stop 5 10

Thp 170 100

TOp 1,5 0,9

Trc 2,4 5

TrUtp 87 75

WarUtp 50 55

It is dicult to establish a period of time to review the integrated model. It depends on

the variability of the market, changes in warehouse capacity (structural and human) or

goals.

In this work the updates are classied as minor or major. The minor revisions are

related to little changes requiring a new optimization to update the scale. The variables

could be:

• the xed warehouse conditions (e.g. number of pallets space, employees, equipments)

The major updates usually require the remodeling of the entire methodology. Some

7.7. Conclusions 147

examples are: changes in indicator equations; modication of variable limits in the optimi-

zation model due to big changes in capacity or process. Moreover, the indicator relation-

ships can change over time, and the manager needs to revise the model when he observes

this tendency.

7.7 Conclusions

The chapter presents the nal integrated performance model with a scale used to analyze

To determine the nal integrated performance model, an analysis of the Jacobian and

correlation results is carried out in order to improve the PCA outcome. The main objectives

are to keep the greatest quantity of indicators as possible with a minimum number of

principal components. The comparison of the worst results obtained from the Jacobian

At the end, seven indicators are eliminated from the analysis and the remain 33 are

designated in six dierent principal components. It is interesting to note that from the

seven indicators, ve are related to quality measures in receiving, storing, replenishment

The six component equations compose the global performance measure. The GP is

optimized to obtain the upper and lower values of the GP scale. The method used to dene

the optimization model can be generalized; however, each warehouse should construct its

own optimization model since it is necessary to dene the variable limits according to the

warehouse reality. The optimized scale, OS, is transformed in a named normal scale, NS,

Finally, the utilization of the aggregated model simulating two dierent warehouse per-

formances is tested. In the rst situation, the outbound indicators have their performance

improved and inbound measures have bad results. For the second test we dene the oppo-

site, outbound indicators have bad results whereas the inbound indicators are great. The

global performance indicator provides better result when outbound indicators are better.

Regarding the exclusion of quality indicators in some warehouse activities (during PCA

analysis) and the result of the test considering dierent indicator results, it might conrm

that the time and productivity are the essential performance axes for the majority of

internal warehouse activities, and the quality level must be guaranteed at the end of the

process chain, with measures related to customer satisfaction. However, this hypothesis

Chapter 8

Conclusions and suggestions for future research

ce n'est que par des lacunes de notre connaissance.

Pierre Simon de Laplace

Contents

8.2 Future Research Directions . . . . . . . . . . . . . . . . . . . . . . . 153

8.2.1 Short-term Research Directions . . . . . . . . . . . . . . . . . . . . . . . 153

8.2.2 Long-term Research Directions . . . . . . . . . . . . . . . . . . . . . . . . 154

Abstract

The chapter is divided in two main sections: rstly, the general conclusions about the

developments carried out throughout this thesis are discussed regarding the objectives

presented in Chapter 1; secondly, research directions are proposed in two dierent

subsections, which split the suggestions by their complexity in short-term and long-

term future researches.

150 8. Conclusions and suggestions for future research

8.1 Conclusions

A dissertation is developed to attain predened objectives. The conclusions serve as a

check out of the accomplishments according to the goals, closing the loop. In the following

items we review the objectives presented in Chapter 1 and discuss the outcomes achieved.

the structured literature review on warehouse performance carried out in this thesis,

the warehouse performance indicators extracted from papers are classied as direct

Even if there is a tendency in the literature to develop indirect measures", they are

not used for daily management since they require a great quantity of data, which

are sometimes dicult to obtain. Therefore, we can conclude that direct indicators

The main insight coming from the literature analysis is that, for the direct indicators,

there is not always a consensus on the denitions of some of the indicators and their

boundaries across the warehouse, resulting in dierent measures for the same metric.

are given, or based on the best common sense if the denitions are not provided.

tors obtained from the literature. In this framework we classify indicators not only

according to quality, cost, time and productivity dimensions, but also in terms of

warehouse activities (receiving, storage, picking, shipping and delivery). The most

frequently used indicators are labor productivity, throughput, on-time delivery, order

lead time and inventory costs. The result of this classication shows that the number

of outbound indicators is much higher than the number of inbound indicators. This is

not very surprising as the warehouse activities are getting more and more customer

oriented. This reveals that the outbound processes/activities are considered more

critical than the inbound ones and hence they are subject to more control.

mance measurement: It consists in four main steps executed to achieve the best

aggregation of the indicator set according to their relationships. The main outcomes

are few (or just one) equation(s) used to measure the global performance with a scale

gated model: the analytical model and the Jacobian matrix measurement to analyze

indicator relationships; the statistical tools to propose indicator groups; the optimi-

zation model to develop the scale for the integrated indicator. This multidisciplinary

Moreover, the methodology can be viewed as general; it gives some alternatives that

one can choose when developing his integrated model. Each warehouse can present

dierent objectives, processes, particularities, and the fact of not specifying all pa-

8.1. Conclusions 151

equations: This is the rst step for the methodology application and it is considered

an outcome of the thesis because usually the performance measurement does not

To apply the methodology, it is necessary to identify the indicator set that will be

ical warehouse, the metric system to assess the standard warehouse performance is

dened, rstly based on the literature review. After some adjustments, a total of 41

indicators compose the metric system, representing all activities that the standard

Even if the analytical model can not be generalized, it could be adapted for some

further models.

The most interesting kind of indicators that are not found in the literature are the

ones related to the replenishment activity. Indeed, we have not found any indica-

our analytical model brings new informations for managers to better evaluate the

warehouse performance.

The use of the Jacobian matrix to identify indicator relationships is one of the most

The Jacobian matrix calculates the partial derivatives of the independent inputs re-

lated to the outputs. To verify the independent inputs of the indicator equations, the

last ones are expanded, creating the data equations. This group of equations builds

the complete analytical model, which describe analytically all relations among data.

tion of the results provided by the Jacobian matrix (indicators x data) permits the

This result is compared with the correlation matrix of indicators. We note that the

majority of indicators with very low correlations corroborate with the indicators shar-

ing the least amount of data in the Jacobian matrix. However, the results are not

conclusive, since there are exceptions and the number of shared data can not dene

the relationship's strength as in correlation matrix. We just verify that the informa-

tions provided by the correlation and the Jacobian seems to be complementary, since

some indicators are maintained in the integrated model having a great quantity of

Finally, we conclude that it is very hard to quantitatively determine from the par-

tial derivatives the intensity of the relationship between indicators. The procedure

described is only used in this thesis to quantify the number of shared data, which

152 8. Conclusions and suggestions for future research

provides a preliminary view of indicator relationships and veries if the results are

grated performance: The literature about scale denition is vast, but it is usually

dened for a unique variable. For instance, the quality and productivity performance

indicators are evaluated using dierent scales. Thus, the development of a scale for

several variables is less common. There are some propositions in the literature to

overcome this issue. In this thesis, we use an optimization approach to obtain the

The optimization model contains the integrated performance model (composed of six

component equations), the analytical model with indicator and constraint equations

and the global performance indicator (which is the aggregation of the components in

one measure). The method used to dene the optimization model can be generalized;

however, each warehouse should construct its own optimization model since it is

The algorithm used to perform the optimization is the SQP, which can handle several

constraints. However, it is very sensitive to the starting values dened for the inputs.

Tests are made to reduce the chances of getting stuck in a local minimum, but other

kinds of tests to verify the results are not done. We believe that this rst optimiza-

tion attained reasonable results regarding the purpose of this thesis, facilitating the

Since the specic objectives are achieved, we conclude that the same happens for the

general one:

The methodology application achieves an integrated model which keeps the majority

represent them. It denotes a very good result, since one of the objectives of this thesis is

to develop a tool that will help managers in the evaluation of a great quantity of informa-

tion. The usability of the integrated model with its scale is tested with indicator values of

two dierent months. In the rst month the outbound indicators have their performance

improved and inbound measures are worst and in the second month is simulated the oppo-

site. The result in the case that outbound indicators are prioritized attains better global

performance.

Finally, we conclude that the methodology proposed in this thesis achieves the objective

of providing insights about indicator relationships, the global warehouse performance and

warehouse activity;

8.2. Future Research Directions 153

4. the development of the complete analytical model with indicator and data equations;

6. the model used to generate the database for the standard warehouse including data

9. the aggregation of several dierent methods (basic statistical tools, partial derivatives

This section is divided in two dierent subsections because we understand that the sugges-

tions presented here have considerable dierences in the development time. The short-term

research directions treat new studies in the warehouse performance subject and possible

applications of the methodology. On the other hand, the long-term research directions

are, in our point of view, new developments that demand more study and time to be

accomplished.

In this section, we basically report some new developments that can be made to improve

• The rst one is the application of the proposed methodology in a real warehouse,

that are not considered in this work as, for example, measures related to reverse

• The SEM (Structural Equation Modeling) method is usually used to verify if a pre-

dened model (i.e. framework dening variable relationships) ts the data. As our

study is exploratory (we did not know how indicators would be aggregated) we did

not use this method in the thesis. However, from the proposed integrated model

the application of SEM using autocorrelated data (e.g. time series) requires special

mathematical manipulations.

tistical tools to compare the indicator relationships obtained with the PCA method.

Among the tools, the DFA theory (Dynamic Factor Analysis) suggest this method

as the best one for our study purpose due to data characteristics. An initial test is

performed in this thesis (Appendix F) but the results are not consistent and reliable,

indicating that more studies should be carried out for DFA utilization.

154 8. Conclusions and suggestions for future research

• The investigation of using the Jacobian matrix to measure strengths between indi-

cator relationships is another point for improvement. The suggestion here is to nd

to the ones of the correlation matrix. One suggestion could be to standardize the

input data and calculate the Jacobian to analyze the relationships, verifying which

are strong or weak. However, to be sure that the results are reliable to dene relation

of constructing the standard Jacobian resides in the input data used to calculate

the partial derivatives. As they come from the time series, which change each new

Warehouses are essential for logistics operations and they have been extensively studied in

the literature. However, the research eort focusing on warehouse performance measure-

ment is not so abundant as for logistics performance. Based on the tendencies identied

in the selected papers, we highlight several future research directions in warehouse man-

agement as follows:

subject, we identify new study tendencies in two main directions: the assessment

tomation inuencing warehouse productivity (De Koster and Balk, 2008)); and the

evaluation of concepts not usually expressed as ratios and, therefore, not measured

• There are dierent types of warehouses. For instance, the manufacturing company

can own the warehouse in which only their products are processed. A warehouse could

products coming from dierent suppliers are treated. Or, a warehouse could be a

retailer's warehouse. In all these cases, the key performance issues can dier since

the goals may dier. Similarly, the management policies within a warehouse may also

aect the way the performance needs to be measured. For instance, for a warehouse

more crucial compared to those which do not implement this technique. One future

research direction is to investigate to what extent the warehouse type inuences the

for analysis. The indicators found in papers usually focus on operational labor.

However, the administrative process has also an important role in the warehouse

performance. For instance, indicators like order lead time and number of perfect

the performance of the warehouse administration is not measured separately and its

impact on the other performance indicators are rarely investigated. This could be

8.2. Future Research Directions 155

• Indicators about reverse logistics have already been developed to evaluate backorder

operations, for example. The productivity and costs of these operations are important

for the enterprise as a whole since they involve customer satisfaction. However, papers

integrating these operations with the main warehouse performance indicators are still

ing emissions and waste indicators with the maximum levels allowed by ISO 14001.

Matopoulos and Bourlakis (2010) go further including indicators of the three pillars

able operations have been widely studied in past years, but the inclusion of metrics

Regarding specically the methodology proposed in this dissertation, other studies can

be suggested.

for strategic areas (e.g. the enterprise performance). One important point, that may be

veried, is the indicators used in the analytical model. Since strategic performance encom-

passes other actors of the supply chain (e.g. suppliers, third party logistics, stakeholders)

besides the focal company, the inclusion of indicators strongly inuenced by external fac-

tors can make the evaluation of performance dicult because it restricts the actions that

Secondly, the generalization of the proposed scale is another point for development. A

companies of the same area and determining the scale from the results obtained. This

development has, for example, huge diculties as the determination of the same analytical

model for all companies (that can compete in the same area but with dierent strategies)

and the denition of the optimization limits due to the diverse situations found among

enterprises.

Finally, we observe, in the last decade, an increasing complexity in the warehouse oper-

IT tools in warehouses and DCs. Since 2000, more complicated algorithms and simulations

(WMS), are recognized as useful means to manage resources in the warehouse (Lam et al.,

2011). The trend of using information systems in warehouse management is a growing ten-

dency and the related new technologies (e.g. augmented reality, RFID, Internet of Things),

will certainly inuence the way the performance is measured and used for decision making

in the future. Therefore, studies regarding the impact and use of these new technologies

Bibliography

Abdi, H., Williams, L. J., and Valentin, D. (2013). Multiple factor analysis: principal

component analysis for multitable and multiblock data sets. Wiley Interdisciplinary

Reviews: Computational Statistics, 5(2):149179.

Andreji¢, M., Bojovi¢, N., and Kilibarda, M. (2013). Benchmarking distribution centres

using Principal Component Analysis and Data Envelopment Analysis: A case study of

Autry, C. W., Gris, S. E., Goldsby, T. J., and Bobbitt, L. M. (2005). Warehouse Manage-

Banaszewska, A., Cruijssen, F., Dullaert, W., and Gerdessen, J. C. (2012). A framework

for measuring eciency levels - The case of express depots. International Journal of

Production Economics, 139(2):484495.

Operations & Production Management, 19(3):275292.

Bentler, P. M. and Chou, C.-P. (1987). Practical Issues in Structural Modeling. Sociological

Methods & Research, 16(1):78117.

Berrah, L., Mauris, G., Haurat, A., and Foulloy, L. (2000). Global vision and performance

225.

Management, 22(2):241264.

Bisenieks, J. and Ozols, E. (2010). The problem of warehouse operation, its improvement

and development in company's logistics system. Human Resources: The Main Factor of

Regional Development, (3):206213.

158 BIBLIOGRAPHY

Böhm, A. C., Leone, H. P., and Henning, P. (2007). Industrial Supply Chains : Performance

Measures , Metrics and Benchmarks. In Plesu, V. and Agachi, P. S., editors, 17th

762.

Bowersox, D. J., Closs, D. J., and Cooper, M. B. (2002). Supply Chain Logistics Manage-

ment. McGraw-Hill, Michigan State University, rst edition.

Cagliano, A. C., DeMarco, A., Rafele, C., and Volpe, S. (2011). Using system dynamics in

Management, 22(2):171188.

Cai, J., Liu, X., Xiao, Z., and Liu, J. (2009). Improving supply chain performance man-

Support Systems, 46(2):512521.

Available at: www.bu.ufsc.br, page 308.

Caplice, C. and She, Y. (1994). A Review and Evaluation of Logistics Metrics. The

International Journal of Logistics Management, 5(2):1128.

Chan, F. T. and Qi, H. (2003). An innovative performance measurement method for supply

2015-03-29).

tural Equation Modeling of Time Series Data. PhD thesis, Hanyang University.

Chow, G., Heaver, T. D., and Henriksson, L. E. (1994). Logistics Performance: Denition

ment, 24(1):1728.

Chow, S.-M., Ho, M.-h. R., Hamaker, E. L., and Dolan, C. V. (2010). Equivalence and Dif-

BIBLIOGRAPHY 159

Clivillé, V., Berrah, L., and Mauris, G. (2007). Quantitative expression and aggregation

tional Journal of Production Economics, 105(1):171189.

Operational Research, 58(1):313.

and Jackknife techniques: construction of condence intervals and tests of hypotheses.

PhD thesis, Pontifícia Universidade Católica do Rio de Janeiro.

183.

Journal of Operations & Production Management, 25(7-8):762780.

De Koster, R., Le-Duc, T., and Roodbergen, K. J. (2007). Design and control of ware-

182(2):481501.

De Marco, A. and Giulio, M. (2011). Relationship between logistic service and maintenance

Dotoli, M., Fanti, M. P., Iacobellis, G., Stecco, G., and Ukovich, W. (2009). Performance

Conference of IEEE Industrial Electronics, pages 43714376, Porto. IEEE.

Ellinger, A. D., Ellinger, A. F., and Keller, S. B. (2003). Supervisory Coaching Behavior,

Enciu, P., Wurtz, F., and Gerbaud, L. (2010). Proposal of a Language for Describing

Dierentiable Sizing Models for Electromagnetic Devices Design. In 14th Biennial IEEE

Conference on Electromagnetic Field Computation - CEFC, pages 12, Chicago.

Federici, A. and Mazzitelli, A. (2005). Dynamic Factor Analysis with STATA. In Italian

STATA User Group meeting, pages 113, Milan, Italy.

160 BIBLIOGRAPHY

Balanced Scorecard. Saraiva, São Paulo.

cations, 13(3):225241.

Franceschini, F., Galetto, M., Maisano, D., and Mastrogiacomo, L. (2008). Properties of

Journal of Productivity and Performance Management, 57(2):137155.

Franceschini, F., Galetto, M., Maisano, D., and Viticchi, L. (2006). The Condition of

Fugate, B. S., Mentzer, J. T., and Stank, T. P. (2010). Logistics Performance: Eciency,

Research, 4(1-2):2838.

Springer, New York, NY, USA.

Goomas, D. T., Smith, S. M., and Ludwig, T. D. (2011). Business activity monitoring:

Real-time group goals and feedback using an overhead scoreboard in a distribution center.

Gu, J., Goetschalckx, M., and McGinnis, L. F. (2007). Research on warehouse operation:

Gu, J. X., Goetschalckx, M., and McGinnis, L. F. (2010). Research on warehouse design

Research, 203(3):539549.

Gunasekaran, A. and Kobu, B. (2007). Performance measures and metrics in logistics and

supply chain management: a review of recent literature (1995 - 2004) for research and

Gunasekaran, A., Marri, H. B., and Menci, F. (1999). Improving the eectiveness of ware-

housing operations : a case study. Industrial Management & Data Systems, 99(8):328

339.

Hasson, C. J. and Heernan, K. S. (2011). Dynamic factor analysis and the exercise

BIBLIOGRAPHY 161

20):pages 46.

Holmes, E. E., Ward, E. J., and Scheuerell, M. D. (2014). Analysis of multivariate time-

series using the MARSS package. Technical report, Northwest Fisheries Science Center,

Structural Equation Modeling, chapter 1, pages 316. Guilford Publications, New York,

NY.

Ilies, L., Turdean, A.-M., and Crisan, E. (2009). Warehouse Performance Measurement -

Jha, D. K., Yorino, N., and Zoka, Y. (2008). Analyzing performance of distribution system

2008

in Nepal and investigating possibility of reorganization of distribution centers. In

Power Technologies, Vols 1-6, number April, pages 13121317, Nanjing, China.

Jiang, J., Chen, H., and Zhang, X. (2009). Index System of Logistics Performance in

2009, pages 28512856.

Johnson, A., Chen, W. C., and McGinnis, L. F. (2010). Large-scale Internet benchmark-

61(3):280286.

Johnson, R. A. and Wichern, D. W. (2002). Factor Analysis and Inference for Structured

tual factors aecting the impact of RFID. Industrial Management and Data Systems,

111(5):714734.

Kassali, R. and Idowu, E. O. (2007). Economics of Onion Storage Systems Under Tropical

https://sites.google.com/site/econometricsacademy/econometrics-models/principal-

component-analysis, (Date Accessed: 2015-03-28):pages 110.

162 BIBLIOGRAPHY

Kennerley, M. and Neely, A. (2002). A framework of the factors aecting the evolution of

Management, 22(11):12221245.

38(3):1827.

Modeling, chapter 1, pages 118. The Guilford Press, New York, NY, 3th edition.

22(3):209218.

Lam, C. H. Y., Choy, K. L., and Chung, S. H. (2011). A decision support system to facilitate

Technology Management, 22(8):972983.

Lao, S. I., Choy, K. L., Ho, G. T. S., Tsim, Y. C., and Lee, C. K. H. (2011). Real-time

inbound decision support system for enhancing the performance of a food warehouse.

Lao, S. I., Choy, K. L., Ho, G. T. S., Tsim, Y. C., Poon, T. C., and Cheng, C. K. (2012). A

real-time food safety management system for receiving operations in distribution centers.

Lauras, M., Marques, G., and Gourc, D. (2010). Towards a multi-dimensional project

Li, J., Sava, A., and Xie, X. (2009). An analytical approach for performance evaluation

of Production Research, 47(2):403414.

Lohman, C., Fortuin, L., and Wouters, M. (2004). Designing a performance measurement

Lu, C.-S. and Yang, C.-C. (2010). Logistics service capabilities and rm performance of

298.

Luo, S.-q., Liu, L., and Shu-quan, L. (2010). Comprehensive Evaluation of Logistics Per-

formance for Agricultural Products Distribution Center. In 2010 2nd International Con-

ference on E-business and Information System Security, pages 14. Ieee.

BIBLIOGRAPHY 163

nal, (winter):2133.

Manikas, I. and Terry, L. A. (2010). A case study assessment of the operational perfor-

mance of a multiple fresh produce distribution centre in the UK. British Food Journal,

112(6):653667.

a primer, chapter 6, pages 7590. Chapman & Hall/ CRC, Boca Raton, Florida, USA,

3rd edition.

Markovits-Somogyi, R., Gecse, G., and Bokor, Z. (2011). Basic eciency measurement

Social and Management Sciences, 19(2):97101.

retail logistics: Findings from an exploratory study. Journal on Chain and Network

Science, 10(3):207218.

Melnyk, S. A., Stewart, D. M., and Swink, M. (2004). Metrics and performance measure-

ment in operations management: dealing with the metrics maze. Journal of Operations

Management, 22(3):209217.

Menachof, D. A., Bourlakis, M. A., and Makios, T. (2009). Order lead-time of grocery

Minitab Inc. (2009). Help of Minitab Statistical Software. Software Minitab, Release 16

for Windows, page www.minitab.com.

Mitro, I. I., Betz, F., Pondy, L. R., and Sagasti, F. (1974). On Managing Science in the

Systems Age: Two Schemas for the Study of Science as a Whole Systems Phenomenon.

Interfaces, 4(3):4658.

Montgomery, D. C. and Runger, G. C. (2003). Applied Statistics and Probability for Engi-

neers. John Wiley & Sons, Inc., New York, NY, USA, 3rd edition.

the last decade and a research agenda for the next. International Journal of Operations

& Production Management, 25(12):12641277.

Neely, A., Gregory, M., and Platts, K. (1995). Performance measurement system design: A

literature review and research agenda. International Journal of Operations & Production

Management, 15(4):80116.

cessed: 2015-03-29):pages 14.

164 BIBLIOGRAPHY

Ng, I., Scharf, K., Progrebna, G., and Maull, R. (2013). Contextual variety, Internet-of-

things and the choice of tailoring over platform: mass customisation strategy in supply

O'Neill, P., Scavarda, A. J., and Zhenhua, Y. (2008). Channel performance in China: a

1(1):2139.

Resource Economics Review, 37(2):288300.

Patel, B., Chaussalet, T., and Millard, P. (2008). Balancing the NHS balanced scorecard!

Accessed: 2015-08-22).

able at: https://onlinecourses.science.psu.edu/stat505/node/63, (Date Accessed:2015-08-

02):pages 12.

Conservation and Recycling, 53(4):175182.

Ramaa, A., Subramanya, K., and Rangaswamy, T. (2012). Impact of Warehouse Man-

54(1):1420.

Rimiene, K. (2008). The design and operation of Warehouse. Economics and Management,

13:136137.

Rodriguez, R. R., Saiz, J. J. A., and Bas, A. O. (2009). Quantitative relationships between

Industry, 60(2):104113.

Rodriguez-Rodriguez, R., Saiz, J. J. A., Bas, A. O., Carot, J. M., and Jabaloyes, J. M.

(2010). Building internal business scenarios based on real data from a performance

Saetta, S., Paolini, L., Tiacci, L., and Altiok, T. (2012). A decomposition approach for

of Production Research, 50(9):23802395.

BIBLIOGRAPHY 165

Sellitto, M. A., Borchardt, M., Pereira, G. M., and Gomes, L. P. (2011). Environmen-

25:291296.

Seuring, S. and Muller, M. (2008). From a literature review to a conceptual framework for

Sohn, S., Han, H., and Jeon, H. (2007). Development of an Air Force Warehouse Logistics

Research, 183(1):148161.

Management: An International Journal, 2(2):5362.

Staudt, T. (2015). Brushless Doubly-Fed Reluctance Machine Modeling, Design and Op-

Suwignjo, P., Bititci, U. S., and Carrie, A. S. (2000). Quantitative models for performance

Journal of Productivity and Performance Management, 53(8):726737.

08-02).

van den Berg, J. and Zijm, W. (1999). Models for warehouse management: Classication

Vascetta, M., Kauppila, P., and Furman, E. (2008). Aggregate indicators in coastal policy

making: Potentials of the trophic index TRIX for sustainable considerations of eutroph-

Voss, M. D., Calantone, R. J., and Keller, S. B. (2005). Internal service quality: Determi-

& Logistics Management, 35(3):161176.

Analysis.

http://www.ic.unicamp.br/wainer/cursos/1s2013/ml/Lecture18_PCA.pdf, (Date

166 BIBLIOGRAPHY

Wang, H., Chen, S., and Xie, Y. (2010). An RFID-based digital warehouse management

system in the tobacco industry: a case study. International Journal of Production Re-

search, 48(9):25132548.

Wang, Y.-F. and Fan, T.-H. (2011). A Bayesian analysis on time series structural equation

and Partial Least Squares for the Job Salience/Job Satisfaction data analysis.

cessed: 2015-04-11):12.

Wu, Y. and Dong, M. (2007). Combining multi-class queueing networks and inventory

International Journal of Advanced Manufacturing Technology, 37(5-6):564575.

Wu, Y.-J. and Hou, J.-L. (2009). A model for employee performance trend analysis of

Yang, L.-r. and Chen, J.-h. (2012). Information Systems Utilization to Improve Distribu-

tion Center Performance : from the Perspective of Task Characteristics and Customers.

Zuur, A. F., Fryer, R. J., Jollie, I. T., Dekker, R., and Beukema, J. J. (2003a). Estimating

common trends in multivariate time series using dynamic factor analysis. Environmetrics,

14(7):665685.

Zuur, A. F., Tuck, I. D., and Bailey, N. (2003b). Dynamic factor analysis to estimate

common trends in sheries time series. Canadian Journal of Fisheries and Aquatic

Sciences, 60(5):542552.

Appendix A

Complete Analytical Model of Performance

Indicators and Data

This section describes the total group of equations creating the complete analytical model.

The division of indicators by their dimensions (time, productivity, cost, quality) are also

used here. Table A.2, Table A.4, Table A.6 and Table A.8 present the data equations on the

right column of the table whereas the indicator equations (Sections 5.2.4, 5.2.5,5.2.6, 5.2.7)

are repeated in the left column. For example, the rst indicator presented in Table A.2 is

P alU

Pnlo

Rect (Equation 5.1), which is measured by the ratio ∆t(Rec)p per Pal Unlo. These

p=1

data are dened in the right side of the table by the Equations A.1 and A.3, respectively.

The denitions of the components inside data equations are showed in Table A.1, A.3,

A.5 and A.7 with the data units in parenthesis, which follow the same logic as presented

for indicator measures. In these tables, just the data from right-side equations are detailed,

indicator names and data which have already been dened in Section 5.2 are not repeated.

Moreover, a data used in several indicator equations have its equation repeated as many

times as necessary. As there are a lot of data denitions in each table, the data is in

There are three distinguish formats in this complete analytical model, which can be

name are in bold, as Rect ; data used in indicator equation are in sans serif style (e.g.

Pal Unlo); the components inside data equation are in slanted style like Prob Rep. In the

cases where the same component is used in indicator equation and in data equation, we

choose to format it in the higher level. For instance, the term Cor Unlo is used as indicator

data in Equation 5.28 and also as data in Equation A.3; so, it is formated in sans serif style.

The time data equations are presented on the right side of Table A.2 and the meaning of

In practice, the total time of an activity is usually acquired by the dierence between

the beginning and the end of the process, independently of the tasks performed inside it.

But in this study, it is necessary to dene time components for relationship analysis. For

that, the time component equations describes the main important tasks performed by each

activity. For example, Equation A.1 details the arrival of a supplier order as: the time used

the inspection time (∆tInsp); the eective time used to unload products (represent by

WEfRec); the queuing time (∆tQueuerec ), which is not a task but exists in practice when

2 A. Complete Analytical Model of Performance Indicators and Data

the total time is obtained. It is important to note that the unit of each detailed task

already represents the total time to perform it in a month, e.g. ∆tInsp is the total time of

The interpretation of the other time equations is similar of the explained for receiving.

Analyzing the time data with the productivity data, we can conclude that terms like

WEfRec constitute the major part of WH Rec, in some cases even attaining the equality.

Data Meaning

index to represent how many hours of the total available labor hours

index to represent how many hours of the total available labor hours

βord =

the employees are dedicated to customer orders administration.

total time for pallet inspection on its arrival or total time for order

∆t(Insp) =

inspection on its dispatch per month (hour/month)

total time that the pallet/order line/order (depending on the ac-

∆tQueuerec , ∆tQueuesto , ∆tQueuerep , ∆tQueuepick , ∆tQueueship ,

∆tQueuedel (hour/month)

total time for other activities/situations not considered in previous

∆t(Others)1−6 =

equation terms per month (hour/month)

Cor Del = number of orders delivered correctly per month (orders/month)

Cor OrdLi Pick = number of order lines picked correctly per month (orderline/month)

Cor OrdLi Ship = number of order lines shipped correctly per month (orderline/month)

Cor Rep =

tory area per month (pallets/month)

Cor Sto = number of pallets stored correctly per month (pallets/month)

Cor Unlo = number of pallets unloaded correctly per month (pallets/month)

HAdmin = H Adminrep , H Adminpick , H Adminship , H Admindel , H Adminorders .

The H Adminorders refers to the total time between the customer order

Prob Del =

(orders/month)

number of order lines with problems during picking activity per month

Prob OrdLi Pick =

(orderline/month)

number of order lines with problems during shipping activity per month

Prob OrdLi Ship =

(orderline/month)

Continued on next page. . .

A.1. Time indicator model 3

Data Meaning

number of pallets with problems in replenishment operation per month

Prob Rep =

(pallets/month)

Prob Sto = number of pallets stored with problems per month (pallets/month)

Prob Unlo = number of pallets unloaded with problems per month (pallets/month)

WEfDel =

(hour/month)

total eective working hours in picking activity per month

WEfPick =

(hour/month)

total eective working hours in receiving activity per month

WEfRec =

(hour/month)

total eective working hours in replenishment activity per month

WEfRep =

(hour/month)

total eective working hours in storage activity per month

WEfSto =

(hour/month)

total eective working hours in shipping activity per month

WEfShip =

(hour/month)

total employee labor hours available for delivery activity per month

WH Del =

(hour/month)

total employee labor hours available for picking activity per month

WH Pick =

(hour/month)

total employee labor hours available for receiving activity per month

WH Rec =

(hour/month)

total employee labor hours available for replenishment activity per

WH Rep =

month (hour/month)

total employee labor hours available for storing activity per month

WH Sto =

(hour/month)

total employee labor hours available for shipping activity per month

WH Ship =

(hour/month)

Table A.2: Time data equation

4

Indicator Equation Data Equations

P alU

Pnlo P alU

Xnlo

∆t(Rec)p ∆t(Rec)p = WEfRec+H Adminrec +∆tQueuerec +∆tInsp1 +∆tOthers1

p=1

Rect = hour

( pallet ) (5.1) p=1

Pal Unlo (A.1)

WEfRec = βrec × WH Rec (A.2)

Pal Unlo = Cor Unlo + Prob Unlo (A.3)

P alSto

P PX

alSto

∆t(Sto)p ∆t(Sto)p = WEfSto + H Adminsto + ∆tQueuesto + ∆tOthers2 (A.4)

p=1

Putt = hour

( pallet ) (5.2) p=1

Pal Sto

WEfSto = βsto × WH Sto (A.5)

Pal Sto = Cor Sto + Prob Sto (A.6)

P alSto

P PX

alSto

∆t(DS)p ∆t(DS)p = ∆t(Rec) + ∆t(Sto) (A.7)

p=1

DSt = hour

( pallet ) (5.3) p=1

Pal Unlo

Pal Unlo = Cor Unlo + Prob Unlo (A.3)

P alM

Poved P alM

Xoved

∆t(Rep)p ∆t(Rep)p = WEfRep+H Adminrep +∆tQueuerep +∆tOthers3 (A.8)

p=1

Rept = hour

( pallet ) (5.4) p=1

Pal Moved

WEfRep = βrep × WH Rep (A.9)

Pal Moved = Cor Rep + Prob Rep (A.10)

Table A.2 continued from previous page

Indicator Equation Data Equations

OrdLiP

P ick OrdLiP

X ick

∆t(Pick)l ∆t(Pick)l = WEfPick + H Adminpick + ∆tQueuepick + ∆tOthers4

l=1 hour

Pickt = ( orderline ) l=1

OrdLi Pick (A.11)

(5.5)

WEfPick = βpick × WH Pick (A.12)

OrdLi Pick = Cor OrdLi Pick + Prob OrdLi Pick (A.13)

OrdLiShip

P OrdLiShip

∆t(Ship)l ∆t(Ship)l = WEfShip+H Adminship +∆tQueueship +∆tInsp2 +∆tOthers5

X

l=1 hour

Shipt = ( orderline ) l=1

OrdLi Ship (A.14)

(5.6)

WEfShip = βship × WH Ship (A.15)

OrdLi Ship = Cor OrdLi Ship + Prob OrdLi Ship (A.16)

OrdDel

P OrdDel

∆t(Del)o ∆t(Del)o = WEfDel + H Admindel + ∆tQueuedel + ∆tOthers6 (A.17)

X

Delt =

o=1 hour

( order ) (5.7) o=1

Ord Del

WEfDel = βdel × WH Del (A.18)

Ord Del = Cor Del + Prob Del (A.19)

OrdDel

P OrdDel

∆t(Ord)o ∆t(Ord)o = ∆t(Pick) + ∆t(Ship) + ∆t(Del) + H Adminord

X

(A.20)

) (5.8)

o=1 hour

OrdLTt = ( order o=1

Ord Del

H Adminord = βord × WH Admin (A.21)

Ord Del = Cor Del + Prob Del (A.19)

5

6 A. Complete Analytical Model of Performance Indicators and Data

The productivity indicators can be classied in two main groups (shown in Table A.4):

indicators related to labor activities (Equation 5.9 - 5.15) and indicators associated with

The rst group of indicators are related to specic activities. As dened in Section

5.2.5, Equations 5.9 - 5.15 have the objective of evaluating the employees' productivity

considering all available time to work, measured as the total hours that the warehouse is

Regarding the number of employees working in a warehouse, usually the employees are

not dedicated to an activity. For example, the warehouse may have all its reception in the

morning. In this case, the manager assigns a lot of people in the receiving dock during

this period and after the activity is nished the employees are designated for another task.

To model this situation, we take into account that the number of employees working in an

activity is the average number of employees that should work all day long to execute the

same task.

The global labor productivity is presented in Equation A.22. We note that the deliv-

ery productivity is not encompassed by Equation A.22, which is limited to the warehouse

boundaries. Even considering in this work the delivery activity as part of warehouse man-

The second group of indicator equations are related to capacity utilization (e.g. ware-

house utilization, Equation 5.19) and global warehouse productivity (represented by Turnover,

Equation 5.17, and Throughput, Equation 5.21). We remark three details about capacity

indicators: (i) it is shown in Equation A.30 that Inv Cap is measured in the number of

pallets available, but depending on the product characteristics other alternative is to use

the unit m3 ; (ii) the inventory capacity used, Inv CapUsed, demonstrated in Equation A.29,

also makes part of the warehouse used areas in Equation A.35, since the inventory area is

an important part of warehouse space. The Inv CapUsed just needs to be transformed to

m2 to stay in accordance with the indicator unit; (iii) the kilograms available, Kg Avail, in

Equation A.34, are calculated in a dynamic way since it considers the number of travels

that a truck can make in a month. Other alternative is to determine the Kg Avail in a

turnover, Equation 5.17, is measured in nancial terms because the data available in the

company are usually in this format. Indeed, the company takes out of the information sys-

tem the data CGoods and Ave Inv ready, without necessity of making calculations. Anyway,

the CGoods and Ave Inv equations are presented in A.31 and A.32, respectively. Analyzing

the Cost of Goods, it makes part of turnover, Equation 5.17, and sales, Equation A.49. A

product is considered sold when it is delivered to the client. So, CGoods is measured by

the number of products delivered times their costs. As the average inventory is dened in

products and not in orders, the number of orders delivered, Ord Del, are also multiplied by

A.2. Productivity indicator model 7

Data Meaning

ave inv = average number of products in inventory (products/month)

area used war =

2

warehouse oor area occupied (m )

Cor Del = number of orders delivered correctly (orders/month)

Cor OrdLi Pick = number of order lines picked correctly per month (orderline/month)

Cor OrdLi Ship = number of order lines shipped correctly per month (orderline/month)

Cor Rep =

area (pallets/month)

Cor Sto = number of pallets stored correctly(pallets/month)

Cor Unlo = number of pallets unloaded correctly (pallets/month)

divided by activity: empl Rec, empl Sto, empl Rep, empl Pick, empl

empl =

Ship. The empl Del is a x number during all available time because

total number of hours during which equipments are stopped per month

HEq Stop =

(hours/month)

total number of hours during which the equipments are working per

HEq Work =

month (hours/month)

total number of hours during which the warehouse operates per day

HWarOperate =

(hours/day)

kg Prod = weight of each product (kg/product)

number of travels made per truck for delivery in a month

nb_travel =

(travel/month)

number of orders with problems during delivery activity

Prob Del =

(order/month)

number of order lines with problems during picking activity

Prob OrdLi Pick =

(orderlines/month)

number of order lines with problems during shipping activity per month

Prob OrdLi Ship =

(orderlines/month)

number of pallets with problems in replenishment operation

Prob Rep =

(nb/month)

Prob Sto = number of pallets stored with problems per month (pallets/month)

Prob Unlo = number of pallets unloaded with problems per month (pallets/month)

Prod Cost =

($/product)

Prod Line = average number of products per order lines (products/orderline)

Prod Ord = average number of products per customer order (products/order)

8 A. Complete Analytical Model of Performance Indicators and Data

Data Meaning

total of products processed by the warehouse per month

Prod Proc =

(products/month)

total number of hours during which the warehouse is open per month

War WH =

(hour/month)

total employee labor hours available for delivery activity per month

WH Del =

(hour/month)

WH Others = sum of employee labor hours working in other activities (hour/month)

total employee labor hours available for picking activity per month

WH Pick =

(hour/month)

total employee labor hours available for receiving activity per month

WH Rec =

(hour/month)

total employee labor hours available for replenishment activity per

WH Rep =

month (hour/month)

total employee labor hours available for storing activity per month

WH Sto =

(hour/month)

total employee labor hours available for shipping activity per month

WH Ship =

(hour/month)

Table A.4: Productivity data equations

Indicator Equation Data Equations

Prod Proc products

Labp = ( hour ) (5.9) Prod Proc = Prod Ship = OrdLi Ship × Prod Line (5.42)

WH

WH = WH Rec + WH Sto + WH Rep + WH Pick + WH Ship + WH Others

(A.22)

Recp = ( ) (5.10) Pal Unlo = Cor Unlo + Prob Unlo (A.3)

WH Rec hour

WH Rec = empl Rec × War WH (A.23)

Pal Sto

Stop = ( pallets ) (5.11) Pal Sto = Cor Sto + Prob Sto (A.6)

hour

WH Sto

WH Sto = empl Sto × War WH (A.24)

Repp = ( hour ) (5.12) Pal Moved = Cor Rep + Prob Rep (A.10)

WH Rep

WH Rep = empl Rep × War WH (A.25)

Pickp = ( hour ) (5.13) OrdLi Pick = Cor OrdLi Pick + Prob OrdLi Pick (A.13)

WH Pick

WH Pick = empl Pick × War WH (A.26)

Shipp = ( hour ) (5.14) OrdLi Ship = Cor OrdLi Ship + Prob OrdLi Ship (A.16)

WH Ship

WH Ship = empl Ship × War WH (A.27)

Continued on next page. . .

9

Table A.4 continued from previous page

10

Indicator Equation Data Equations

Ord Del order

Delp = ( ) (5.15) Ord Del = Cor Del + Prob Del (A.19)

WH Del hour

WH Del = empl Del × War WH (A.28)

Inv CapUsed n

× 100(%) (5.16) ave invi

P

InvUtp =

Inv Cap

Inv CapUsed =

i=1

(A.29)

Prod pal

i = 1, . . . , n = SKU's

Inv Cap = total amount of pallet space (A.30)

CGoods n

(5.17) Del × Prod Ord)i × Prod costi )

X

TOp =

Ave Inv

(times) CGoods = ((Ord (A.31)

i=1

n

(ave invi × Prod costi )

X

Ave Inv = (A.32)

i=1

i = 1, . . . , n = SKU's

Kg Tr n

TrUtp = × 100(%) (5.18) Del × Prod Ord)i × kg Prodi

X

Kg Avail Kg Tr = (Ord (A.33)

i=1

i = 1, . . . , n = SKU's

m

capa × nb_travela

X

Kg Avail = (A.34)

a=1

a = 1, . . . , m = number of trucks

Continued on next page. . .

Table A.4 continued from previous page

Indicator Equation Data Equations

War CapUsed warXarea

WarUtp =

War Cap

× 100(%) War CapUsed = war used area (A.35)

(5.19) b=1

War Cap = total useful warehouse area (A.36)

HEq Stop z z

EqDp = × 100(%) (5.20) HEq Stopc + HEq Workc

X X

HEq Avail HEq Avail = HEq Stop + HEq Work =

c=1 c=1

(A.37)

c = 1, . . . , z = nb of equipments

Thp = ( hour ) (5.21) Prod Ship = OrdLi Ship × Prod Line (A.38)

War WH

War WH = HWarOperate × days month (A.39)

11

12 A. Complete Analytical Model of Performance Indicators and Data

The cost equations are presented in Table A.6 whereas their denitions are in Table A.5.

The distribution costs (Equation 5.23) are measured but not included in the total

warehouse costs (Equation A.48). The salary costs of delivery employees are also included

in Equation 5.23, instead of being considered in labor cost indicator (Equation 5.26).

Regarding the labor cost indicator (Equation 5.26), only the employees working inside

the warehouse are taken into account. The time of the administrative employees are divided

in: hours dedicated to customer orders and hours dedicated to other warehouse activities.

The rst part, hours dedicated to customer orders, are included in order processing costs

(Equation 5.24), and the second part, hours dedicated to other warehouse activities, are

included in the labor cost (Equation 5.26). When the total warehouse costs (Equation A.48)

are assessed, order processing cost and labor cost are summed up, and the administrative

should be interpreted as the quantity of prot lost due to the absence of inventory to

fulll customer orders. The lack of stock is measured by the quality indicator stock out

(Equation 5.40). This percentage of missing stock is multiplied by the total products

picked in a month (named Prod Out, Equation A.47) and the average prot gain with each

product sold.

A.3. Cost indicator model 13

Data Meaning

index representing the partial quantity over the Salary payed as Charges per

α=

month

index to represent how many hours of the total available labor hours the em-

βord =

ployees are dedicated to customer orders administration.

$ oil = oil price per liter ($/l)

cost per hour worked in each activity. It is divided by activities: $/hrec , $/hsto ,

$/h =

$/hrep , $/hpick , $/hship , $/hdel , $/hadmin , $/hother ($/hour)

Ave Inv = average inventory in warehouse ($/month)

CGoods = total cost of items sold ($)

Cor Del = number of orders delivered correctly per month (orders/month)

deprec 1−2 = depreciation costs of company assets used in activities per month ($/month)

l_used = mean of oil liters used by trucks for one travel (liter/travel)

nb_travel = number of travels made per truck for delivery in a month (travel/month)

Other1−2 = other costs not considered in equation ($/month)

Prob Del = number of orders with problems during delivery activity (orders/month)

Prod Cost = cost of products arriving in warehouse, the purchasing price ($/product)

Prod Out = number of products taken out of the inventory (products/month)

Prot = average gross prot of products sold ($/product)

Rate = monthly nancial rate (%)

Charges tr = Labor charges payed over salary value ($/month)

SL = service level oered to the customer (%)

Salary tr = total salaries of delivery employees per month ($/month)

Truck MaintC = total cost of truck maintenance ($/month)

WH Admin = total employee labor hours available in administration activity (hour/month)

total employee labor hours available for delivery activity per month

WH Del =

(hour/month)

WH Others = sum of employee labor hours working in other activities (hour/month)

total employee labor hours available for picking activity per month

WH Pick =

(hour/month)

total employee labor hours available for receiving activity per month

WH Rec =

(hour/month)

total employee labor hours available for replenishment activity per month

WH Rep =

(hour/month)

total employee labor hours available for storing activity per month

WH Sto =

(hour/month)

total employee labor hours available for shipping activity per month

WH Ship =

(hour/month)

Table A.6: Cost data equations

14

Indicator Equation Data Equations

n

(ave invi × Prod costi )

X

Ave Inv = (A.32)

i=1

LostC = (1 − SL) × Prot × Prod Out (A.41)

StockOutq )

SL = 1 − ( (A.42)

100

+ Salarytr + Chargestr + deprec1 + Other1 (A.43)

TrC

Trc = $

( order ) (5.23)

Ord Del Salarytr = $/hdel × WH Del (A.44)

Chargestr = α × Salarytr and 0<α<1 (A.45)

Ord Del = Cor Del + Prob Del (A.19)

Ord ProcC

OrdProcc = $

( order ) (5.24) + deprec2 + Other2 (A.46)

Cust Ord

Continued on next page. . .

Table A.6 continued from previous page

Indicator Equation Data Equations

War Cost

Sales = CGoods + (Prot × Ord Del × Prod Ord) (A.49)

CSc = (%) (5.25) n

Sales

Del × Prod Ord)i × Prod costi )

X

CGoods = ((Ord (A.31)

i=1

i = 1, . . . , n = SKU's

Labc = Salary + Charges + Others( month )

(5.26) + $/hother × WH Others (A.50)

$

Maintc = BuildC+EqMaintC+Others( month ) BuildC = building maintenance costs (A.51)

(5.27) EqMaintC = maintenance cost of all equipments (A.52)

15

16 A. Complete Analytical Model of Performance Indicators and Data

The expressions of the quality problems presented in Equations A.53 - A.59 are inequalities.

The objective of these expressions is to show the main data shared by dierent quality

indicators. For example, the number of order lines picked with problem (Equation A.57)

contain as the main errors: scraps, data error and order lines no available. The problems

represented by scrap and items no available are also used in Scrapq and StockOutq

quality indicators, respectively.

Regarding the inequality result, the total number of order lines picked with problem is

equal or smaller than the sum of problems since in a real situation an order line can have

more than one problem at the same time. The correct orders are the ones with no problem

An important consideration about the scraps inserted in indicator equations is that they

do not impact the nal number of orders processed. It is determined that these scraps are

the ones that have been replenished during the same month. As this situation can happen

in practice (scraps not replenished in the same month), we include scraps not solved in the

The new terms introduced in the right side of Table A.8 are presented in Table A.7.

Data Meaning

Cor Del = number of orders delivered correctly per month (orders/month)

Cor OrdLi Pick = number of order lines picked correctly per month (orderline/month)

Cor OrdLi Ship = number of order lines shipped correctly per month (orderline/month)

number of pallets moved correctly from the reserve storage to the forward

Cor Rep =

picking area per month (pallets/month)

Cor Sto = number of pallets stored correctly per month (pallets/month)

Cor Unlo = number of pallets unloaded correctly per month (pallets/month)

number of products with data system errors from outbound area per

data error =

month (products/month)

number of pallets with data system errors from the activities: unload-

ing, storing and replenishment. It is the complement of cor data in

error data system1−3 =

system (orders/month), and the sum of all errors result in Prob Data

(orders/month)

number of orders shipped incomplete on rst shipment per month

NoComplet Ord Ship =

(orders/month)

Ord Ship = number of orders shipped per month (orders/month)

others = number of other problems not dened per month (nb/month)

number of orders with delays per month. The opposite of order on time

ord late =

(orders/month)

number of order lines per month that are not available in stock when

OrdLi noAvail =

the customer makes an order (orderlines/month)

number of pallets with inaccuracies between the physical inventory and

Prob data =

the system per month (pallets/month)

number of orders with problems during delivery activity per month

Prob Del =

(orders/month)

A.4. Quality indicator model 17

Data Meaning

Prod Line = average number of products per order lines (products/orderline)

number of order lines with problems during picking activity per month

Prob OrdLi Pick =

(orderline/month)

number of order lines with problems during shipping activity per month

Prob OrdLi Ship =

(orderline/month)

Prod pal = average number of products stocked per pallet (products/pallet)

number of products processed by the warehouse per month. Products

Prod Proc = processed refers to the number of products shipped in the warehouse

(products/month)

number of pallets with problems in replenishment operation

Prob Rep =

(pallets/month)

Prob Sto = number of pallets stored with problems per month (pallets/month)

Prob Unlo = number of pallets unloaded with problems per month (pallets/month)

number of products per month that are not available in stock when the

Prod noAvail=

customer makes an order (product/month)

Prod Ord = average number of products per customer orders (products/order)

number of products taken out of the inventory per month

Prod Out =

(products/month)

number of pallets with losses from handling problems or accidents per

scrap1−3 = month (pallets/month). scrap 4−5 has the same meaning, it is just mea-

sured by (orderlines/month). scrap 6 is measured in (orders/month)

Table A.8: Quality data equations

18

Indicator Equation Data Equations

Recq = × 100(%) (5.28)

Pal Unlo

Prob Unlo 6 scrap1 + error data system1 + others (A.53)

Cor Sto Pal Sto = Cor Sto + Prob Sto (A.6)

Stoq = × 100(%) (5.29)

Pal Sto

Prob Sto 6 scrap2 + error data system2 + others (A.54)

Repq = × 100(%) (5.30)

Pal Moved

Prob Rep 6 scrap3 + error data system3 + others (A.55)

Pal Unlo + Pal Sto + Pal Moved - Prob data

Invq = ×100 Pal Moved = Cor Rep + Prob Rep (A.10)

Pal Unlo + Pal Sto + Pal Moved

(5.31) 3

error data systemm

X

Prob Data = (A.56)

m=1

Cor OrdLi Pick

Pickq = × 100(%) (5.32)

OrdLi Pick Prob OrdLi Pick 6 scrap4 + data error + OrdLi noAvail + others

(A.57)

Continued on next page. . .

Table A.8 continued from previous page

Indicator Equation Data Equations

Cor OrdLi Ship

Shipq = × 100(%) (5.33)

OrdLi Ship Prob OrdLi Ship 6 scrap5 +data error+No OT Ship+NoComplet Ord Ship+others

(A.58)

Cor Del

Delq = × 100(%) (5.34)

Ord Del Prob Del 6 scrap6 + data error + ord late + no complete ord + others

(A.59)

OTDelq = × 100(%) (5.35)

Ord Del

Ord Del = Cor Del + Prob Del (A.19)

Ord Ship OT

OTShipq = × 100(%) (5.36) OrdLi

Ord Ship XShip

Ord Ship = OrdLi Shipp (A.62)

p=1

Complet 1st Ship

OrdFq = × 100(%) (5.37) OrdLi

Ord Ship XShip

Ord Ship = OrdLi Shipp (A.62)

p=1

Continued on next page. . .

19

Table A.8 continued from previous page

20

Indicator Equation Data Equations

(Ord OT, ND, CD) Ord OT, ND, CD = orders on time, with no damages and correct documents

PerfOrdq = × 100(%) (5.38)

Ord Del

Ord Del = Cor Del + Prob Del (A.19)

Ord Del - Cust Complain Cust Complain = customer complaints regarding warehouse processes

CustSatq = × 100(%) (A.64)

Ord Del

(5.39) Ord Del = Cor Del + Prob Del (A.19)

StockOutq = × 100(%) (5.40)

Prod Out

Prod Out = Ord LiPick × Prod Line (A.47)

Nb Scrap

Scrapq = × 100(%) (5.41) (scrap4 + scrap5 ) × Prod Line + scrap6 × Prod Ord (A.65)

Prod Proc

Appendix B

Data Generation

This appendix details how data is created for the standard warehouse. The next sections

present separately the product ow and data equations for warehouse operations, demon-

The receiving activity is detailed in Figure B.1, which is divided in ve parts: four rectan-

gles with data equations and one activity ow schema in the up right side of the gure. The

four rectangles shows, respectively: the Global variables; the internal inputs named `IntIn-

put'; the `Outputs' and internal outputs `IntOutput'; the Number of problems occurred

nb_days/month[t] = RANDBETWEEN (20; 25)

HWarOperate = 8 h/day

Supplier Ord [t] = NORM.INV(pr();28;2) +

nb pallets/truck = 25

+ Cor Unlo +

[t-1] +

IntInput – Internal Inputs

Prob Unlo

nb PalRec[t] = Supplier Ord [t] * nb pallets/truck

WHRec[t] = emplRec * HWarOperate * nb_days/month[t]

emplRec = 0,5 IntOutput

Insp time[t] = NORM.INV(pr(); 0,5; 0,1)

Scrap Unlo1[t]

β_rec = 0,85

Outputs

Cor Unlo[t] = RANDBETWEEN ((nb PalRec[t] + Scrap Unlo1[t-1] )*0,98; (nb PalRec[t] + Scrap Unlo1[t-1] )*1)

Prob Unlo[t] = nb PalRec[t] + Scrap Unlo1[t-1] – Cor Unlo[t]

Scrap Unlo1[t] = RANDBETWEEN (0; Prob Unlo[t])

∆t Insp[t] = Insp time[t] * Supplier Ord[t]

WEfRec[t] = β_rec * WHRec[t]

∆t Admin_rec[t] = 1h/day * nb_days/month [t]

Scrap Unlo[t] = RANDBETWEEN (Scrap Unlo1[t]; Prob Unlo[t])

Error DataInb1[t] = RANDBETWEEN (0; Prob Unlo [t]– Scrap Unlo[t])

Other Errors rec = Prob Unlo [t] – Scrap Unlo[t] – Error DataInb1[t]

The Global variables are general information that can be used in any part of the ware-

house to calculate other data or indicators. The number of days worked in a month

`nb_days/month', for example, varies every month between 20 and 25 days, following a

22 B. Data Generation

uniform distribution of probabilities. Once the number of days is dened for a month,

this information is used for all data and indicators in that month. To simplify the gure,

we illustrate only the global variables related to receiving operation and used to calculate

inputs or outputs.

The internal inputs `IntInput' and internal outputs `IntOutput' comprehend data re-

The `Outputs' are also data used on performance indicators, but the dierence is that

these outputs are also the inputs of the next activity, demonstrating the product ow in the

warehouse, which also impact indicator interactions. Finally, the rectangle on the bottom

of Figure B.1 demonstrates the total `Number of problems occurred during the month'.

These data are a sum of all problems occurred during the month in the activity (solved or

not), and some of these informations are also utilized in indicator equations.

The design of Figure B.1 and the information inside rectangles are used as standard

for all other warehouse activities presented in next sections. Moreover, the notation of the

equations inside the rectangles are the same presented in the complete analytical model

described in Appendix A.

In the receiving ow schema of Figure B.1, the number of supplier orders `Supplier Ord'

arriving in the warehouse are a random number varying according to a normal distribution

with mean 28 and standard deviation 2. As the performance indicators in receiving are

measured in pallets, we assess the number of pallets received, `nb PalRec[t]' (rst equation

of `IntInput'), multiplying the number of supplier orders received in the month t and

the number of pallets per truck, `nb pallets/truck'. This equation demonstrates that we

consider all supplier orders arriving with the same quantity, a complete truck of 10 tons

The number of labor hours available to work in a month, `WHRec[t]', change according

to the working days and the number of employees performing the activity. As stated

before, in this scenario, the number of employees are considered constant over time for all

activities.

The last two `IntInput' equations correspond to the time to perform product quality

inspections, Insp time[t] and βord is the index to represent how many hours of the total

available labor hours the employees are eectively receiving. The `Insp time' uses the

normal function to dene the time, in hours, taken by administrative employees to perform

inspection, which is dened as 30 min (0.5 hour) on average for each supplier order with a

standard deviation of 6 minutes (0.1 hour). Insp time[t] and βord are used to calculate the

total inspection time and eective hours receiving in the month [t], named ∆t Insp[t] and

WEfRec[t], respectively. The equations are showed in the `IntOutput' area of the Figure

B.1.

The last formula of IntOutput is ∆t Admin_rec[t] which means the time taken by

administrative personnel to execute activities related to receiving and supplier orders. This

The type of receiving `problems' occurred in a month are not exhaustively detailed.

B.2. Storage data 23

The Scrap Unlo[t] and Error DataInb1[t] are demonstrated separately because their values

are used in ScrapRateq and Invq indicators, respectively. All other possible errors are

identied in equation `Other Errors rec', besides its value is not used for indicator mea-

surement. It is important to note that according to the equations, the number of Error

DataInb1[t] has as limit the number of problems minus products with scrap problems.

Thus, another constraint of the model is not allowing an order with two dierent errors at

The outputs of receiving, Cor Unlo[t] and Prob Unlo[t], variates every month between 98

% to 100% of the total pallets unloaded for Cor Unlo[t] and of 0% up to 2% for Prob Unlo[t].

According to Figure B.1, the Cor Unlo[t] is measured using a uniform random probability

between 98% and 100% of the total inputs, which are the total of pallets received, nb

PalRec[t], and the number of scraps not solved in the previous month [t-1] (Scrap Unlo1[t-

1]). The Prob Unlo[t], in contrast, is calculated just with the dierence between the total of

inputs and the pallets unloaded correctly, Cor Unlo[t]. Therefore, the inputs of the storage

activity (presented in the next section) are the resultant of CorU nlo[t] + P robU nlo[t] −

ScrapU nlo1[t] equation.

All other activities have their equations developed based on the same logic presented

here for the receiving activity. Thus, just particularities not discussed yet are presented in

next sections.

The data equations used in storage activity are presented in Figure B.2.

In storage activity, the outputs Cor Sto[t] and Prob Sto[t], variates every month between

96 % to 98% of the total pallets stored for Cor Sto[t] and of 0% up to 2% for Prob Sto[t]. It

results, in some months, that a number of products could be not all processed, remaining

as Sto in Process for the next month. The Sto in Process is the sum of products

with problems not solved (information arrow getting out of Prob Sto and entering in `No

Proc') with products not processed `No Proc'. It is interesting to note that the problems

not solved are the number of scraps not replaced during the month, represented by Scrap

Sto1[t].

The data equations used in replenishment activity are shown in Figure B.3. The replen-

ishment activity consist on the movement of pallets from the reserve storage area to the

forward picking area. As this activity aims to replenish the inventory picking area, the

number of pallets to move depends on the quantity of products picked (represented by Cor

We note that the replenishment indicators are measured by pallets and the Cor Pick[t]

and Prob Pick[t] have order lines as units. Thus, the equations presented in Figure B.3

24 B. Data Generation

Storage

Global variables

nb_days/month[t] = RANDBETWEEN (20; 25) Cor Unlo[t] + Prob Unlo[t] – Scrap Unlo1[t] +

Cor Sto +

HWarOperate = 8 h/day +

+

[t-1] Prob Sto

No Proc +

IntInput – Internal Inputs +

WHSto[t] = emplSto * HWarOperate * nb_days/month[t]

IntOutput

emplSto = 0,5

β_sto = 0,85

Sto inProcess

Outputs

Cor Sto[t] = RANDBETWEEN ((Sto inProcess[t-1] + Cor Unlo[t] + Prob Unlo[t] – Scrap Unlo1[t])*0,96; (Sto inProcess[t-1] + Cor

Unlo[t] + Prob Unlo[t] – Scrap Unlo1[t])*0,98)

Prob Sto[t] = RANDBETWEEN (0; (Sto inProcess[t-1] + Cor Unlo[t] + Prob Unlo[t] – Scrap Unlo1[t])*0,02)

Sto inProcess[t] = Sto inProcess[t-1] + Cor Unlo[t] + Prob Unlo[t] - Scrap Unlo1[t] – Cor Sto[t] – Prob Sto[t] + Scrap Sto1[t]

Scrap Sto1[t] = RANDBETWEEN (0; Prob Sto[t])

WEfSto[t] = β_sto * WHSto [t]

∆t Admin_sto[t] = 1h/day * nb_days/month [t]

Scrap Sto[t] = RANDBETWEEN (Scrap Sto1[t]; Prob Sto[t])

Error DataInb2[t] = RANDBETWEEN (0; Prob Sto[t] – Scrap Sto[t])

Other Errors sto = Prob Sto[t] – Scrap Sto[t] – Error DataInb2[t]

nb_days/month[t] = RANDBETWEEN (20; 25) Cor Pick[t] + Prob Pick[t]

HWarOperate = 8 h/day

Cor Rep +

Prod_Ord[t] = NORM.INV(pr(); 20; 2)

+

nb prod pal = 40 +

+ Prob Rep

[t-1]

IntInput No Proc +

IntOutput

emplRep = 1

β_rep = 0,8 Rep inProcess[t]

Outputs

Cor Rep[t] = RANDBETWEEN ((((Cor Pick[t] + Prob Pick[t]) *Prod_Ord[t]/ nb products pal)+ Rep inProcess[t-1] ) *0,96; (((Cor

Pick[t] + Prob Pick[t]) *Prod_Ord[t]/ nb products pal)+ Rep inProcess[t-1] )*0,98)

Prob Rep[t] = RANDBETWEEN (0; (((Cor Pick[t] + Prob Pick[t]) *Prod_Ord[t]/ nb products pal)+ Rep inProcess[t-1] )*0,02)

Rep inProcess[t] = ((Cor Pick[t] + Prob Pick[t]) *Prod_Ord[t]/ nb products pal) - Cor Rep[t] - Prob Rep[t] + Scrap Rep1[t] + Rep

inProcess[t-1]

IntOutput

Scrap Rep1[t] = RANDBETWEEN (0; Prob Rep[t])

WEfRep[t] = β_rep * WHRep[t]

∆t Admin_rep[t] = 1h/day * nb_days/month [t]

Scrap Rep[t] = RANDBETWEEN (Scrap Rep1[t]; Prob Rep[t])

Error DataInb3 [t]= RANDBETWEEN (0; Prob Rep[t] – Scrap Rep[t])

Other Errors rep = Prob Rep[t] – Scrap Rep[t] – Error DataInb3[t]

B.4. Picking data 25

The data equations used in picking activity are depicted in Figure B.4.

nb_days/month[t] = RANDBETWEEN (20; 25)

HWarOperate = 8 h/day + Cor Pick +

Prod_Ord[t] = NORM.INV(pr(); 20; 2) +

Demand[t] = NORM.INV(pr(); 28000; 2000) + Prob Pick

[t-1]

IntInput No Proc

Scrap Del[t-1]

WHPick[t] = emplPick*HWarOperate*nb_days/month[t]

IntOutput

emplPick = 4

β_pick = 0,95 Pick inProcess

Outputs

Cor Pick[t] = RANDBETWEEN ((Scrap Del1[t-1] + Pick inProcess[t-1] + CustOrd[t])*0,96; (Scrap Del1[t-1] + Pick inProcess[t-1] +

CustOrd[t])*0,98)

Prob Pick[t] = RANDBETWEEN (0; (Scrap Del1[t-1] + Pick inProcess[t-1] + CustOrd[t])*0,02)

Pick inProcess [t] = Scrap Del1[t-1] + Pick inProcess[t-1] + Cust Ord[t]- Cor Pick[t] - Prob Pick[t] + Scrap Pick1[t] + ProdnoAvail1[t]

IntOutput

Scrap Pick1[t] = RANDBETWEEN (0; Prob Pick[t])

ItemnoAvail1[t] = RANDBETWEEN (0; Prob Pick[t] – Scrap Pick1[t])

WEfPick[t] = β_pick * WHPick[t]

∆t Admin_pick[t] = 1h/day * nb_days/month [t]

Scrap Pick[t] = RANDBETWEEN (Scrap Pick1[t]; Prob Pick[t] – ProdnoAvail1[t])

ProdnoAvail[t] = RANDBETWEEN (ProdnoAvail1[t]; Prob Pick[t] – Scrap Pick[t])

Other Errors pick = Prob Pick[t] – Scrap Pick[t] – ProdnoAvail[t]

Figure B.5 presents the shipping activity with its equations. The indicator Order Fill rate

(Equation 5.37) measures the number of orders delivered complete. Instead of generat-

ing the number of complete orders, we evaluate the number of partial orders delivered,

Figure B.6 shows the delivery activity with its equations.

26 B. Data Generation

nb_days/month[t] = RANDBETWEEN (20; 25)

HWarOperate = 8 h/day Cor Ship +

Cor Pick[t] + Prob Pick[t] – ScrapPick1[t] +

+

– ProdnoAvail1[t] + Prob Ship

[t-1]

No Proc

IntInput

WHShip[t] = emplShip*HWarOperate*nb_days/month IntOutput

emplShip = 3

β_ship = 0,95 Ship inProcess

Outputs

Cor Ship[t] = RANDBETWEEN ((Ship inProcess[t-1] + Cor Pick[t] + Prob Pick[t] – ScrapPick1[t] – ProdnoAvail1[t]) *0,96; (Ship

inProcess[t-1] + Cor Pick[t] + Prob Pick[t] – ScrapPick1[t] – ProdnoAvail1[t]) *0,98)

Prob Ship[t] = RANDBETWEEN (0; (Ship inProcess[t-1] + Cor Pick[t] + Prob Pick[t] – ScrapPick1[t] – ProdnoAvail1[t]) *0,02)

Ship inProcess [t] = Ship inProcess[t-1] + Cor Pick[t] + Prob Pick[t] - Scrap Pick1[t] - ProdnoAvail1[t] - Cor Ship[t] – Prob Ship[t] +

Scrap Ship1[t]

IntOutput

Scrap Ship1[t] = RANDBETWEEN (0; Prob Ship[t])

WEfShip[t] = β_ship * WHShip[t]

∆t Admin_ship[t] = 1h/day * nb_days/month[t]

Scrap Ship[t] = RANDBETWEEN (Scrap Ship1[t]; Prob Ship[t])

NoComplet_Ord Ship[t] = RANDBETWEEN(0; Prob Ship[t] – Scrap Ship[t])

Other Errors ship = Prob Ship[t] – Scrap Ship[t] – NoComplet_Ord Ship[t]

OTShip[t] = RANDBETWEEN (Cor Ship[t]; Cor Ship[t] + Prob Ship[t] – Scrap Ship1[t])

Global variables

Delivery

nb_days/month[t] = RANDBETWEEN (20; 25)

HWarOperate = 8 h/day +

Cor Ship[t] + Prob Ship[t] – ScrapShip1[t] Cor Del +

+

+

Prob Del

IntInput -

WHDel[t] = emplDel*HWarOperate*nb_days/month(t] IntOutput

emplDel = 2

Scrap Del[t]

β_del = 0,90

Outputs

Cor Del[t] = RANDBETWEEN ((Cor Ship[t] + Prob Ship[t] – ScrapShip1[t] )*0,98; (Cor Ship[t] + Prob Ship[t] – ScrapShip1[t])*1)

Prob Del[t] = Cor Ship[t] + Prob Ship[t] – ScrapShip1[t] – Cor Del[t]

Scrap Del1[t] = RANDBETWEEN (0; Prob Del[t])

IntOutput

WEfDel[t] = β_del * WHDel[t]

∆t Admin_del [t] = 2h/day * nb_days/month[t]

Scrap Del[t] = RANDBETWEEN (Scrap Del1[t]; Prob Del[t])

Cust Complain[t] = RANDBETWEEN(0; Prob Del[t])

Other Errors del = Prob Del[t] – Scrap Ship

OTDel[t] = RANDBETWEEN (Cor Del[t]; Cor Del[t] + Prob Del[t] – Scrap Del1[t])

OT_ND_DC[t] = Cor Del[t]

B.7. Warehouse and Inventory data 27

This section demonstrates the equations related to the warehouse as a whole (Figure B.7),

The warehouse building and the truck make part of company assets; it means that all

costs associated with their maintenance are taken into account in cost indicators.

The charges, total paid over salary for all employees are considered as 50% of salary

value. The average of liters used per travel is 2, considering that each travel has 10 km and

5 km is made with one oil liter. The depreciations (deprec1 and deprec2) are considered

nb_days/month[t] = RANDBETWEEN (20; 25)

HWarOperate = 8 h/day

Supplier Ord [t] = NORM.INV(pr();28;2) +

nb pallets/truck = 25

+ Cor Unlo +

[t-1] +

IntInput – Internal Inputs

Prob Unlo

nb PalRec[t] = Supplier Ord [t] * nb pallets/truck

WHRec[t] = emplRec * HWarOperate * nb_days/month[t]

emplRec = 0,5 IntOutput

Insp time[t] = NORM.INV(pr(); 0,5; 0,1)

Scrap Unlo1[t]

β_rec = 0,85

Outputs

Cor Unlo[t] = RANDBETWEEN ((nb PalRec[t] + Scrap Unlo1[t-1] )*0,98; (nb PalRec[t] + Scrap Unlo1[t-1] )*1)

Prob Unlo[t] = nb PalRec[t] + Scrap Unlo1[t-1] – Cor Unlo[t]

Scrap Unlo1[t] = RANDBETWEEN (0; Prob Unlo[t])

∆t Insp[t] = Insp time[t] * Supplier Ord[t]

WEfRec[t] = β_rec * WHRec[t]

∆t Admin_rec[t] = 1h/day * nb_days/month [t]

Scrap Unlo[t] = RANDBETWEEN (Scrap Unlo1[t]; Prob Unlo[t])

Error DataInb1[t] = RANDBETWEEN (0; Prob Unlo [t]– Scrap Unlo[t])

Other Errors rec = Prob Unlo [t] – Scrap Unlo[t] – Error DataInb1[t]

Figure B.8 shows in IntOutput rectangle the equations inv_end[t] and aveinv. The

equation inv_end[t] means the inventory on hand at the end of a given period. It is

calculated by: the inventory from the previous period (inv_end[t-1]), summed up with the

products get in stock (CorSto[t] + ProbSto[t] - ScrapSto1[t]), less the demand in the given

period (CorPick[t] + ProbPick[t]). To calculate the average stock during an entire month,

a data used in some indicators, the equation aveinv is applied for this purpose.

28 B. Data Generation

Global variables

nb_days/month[t] = RANDBETWEEN

(20; 25)

HWarOperate = 8 h/day

Rate = 10% month Inventory

nb prod/pal = 40

palSpace = 1000

Cor Sto[t] + Prob Sto[t] – ScrapSto1[t] Cor Pick[t] + Prob Pick[t]

$/hour admin = 7

deprec2 = $200/ month

IntInput:

Cust Ord[t] = Demand[t] / Prod_Ord[t]

WHAdmin[t] = emplAdmin*HWarOperate*nb_days/month[t] IntOutput

emplAdmin = 2

β_ord = 0,55

Prod_Ord[t] = NORM.INV(pr(); 20; 2)

IntOutput:

inv_end[t] = inv_end[t-1] + (Cor Sto[t] + Prob Sto[t] – ScrapSto1[t])* nb products/pal – (CorPick[t] + Prob Pick[t])*Prod_Ord[t]

aveinv = (inv_end[t-1] + inv_end[t]) / 2

Ord Procc[t] = (WHAdmin[t] * β_ord * $7/hour + 0,5 * (WHAdmin[t] * β_ord * $7/hour) + deprec2)/ Cust Ord[t]

HAdmin_ord = β_ord * WHAdmin[t]

ProdnoAvail[t]= ProdnoAvail1[t]*Prod_Ord[t] + IF(inv_end[t-1] + (Cor Sto[t] + Prob Sto[t] – Scrap Sto1[t])*nb prod /pal –

(Cor Pick[t] + Prob Pick[t] – Scrap Pick[t])*Prod_Ord[t]) > 0; 0; ABS(Cor Sto[t] + Prob Sto[t] – Scrap Sto1[t])*nb prod/pal –

(Cor Pick[t] + Prob Pick[t] – Scrap Pick[t])* Prod_Ord[t]

Appendix C

Manual Procedure to determine indicator

relationships

This appendix demonstrate the initial analysis performed to determine indicator relation-

ships manually.

Initially, we construct a schema (Figure C.1) showing the all 40 indicators and the main

data used to measure them (data from indicator equations of Sections 5.2.4, 5.2.5, 5.2.6,

5.2.7). The indicators are represented by ellipses and data by rectangular blocks. The

lines represent the connection between data and the indicator. For example, the indicator

EqDp (in the up left corner of Figure C.1) is calculated by HEq Stop per HEq Avail (the

green rectangles), so there are lines connecting both data with the indicator EqDp .

HEq Avail HEq Stop War Cap WarUtp War CapUsed Inv Cap Prod Out

EqDp

Recp WH Rec InvUtp

EqMaintC Ave Inv

Rect StockOutq

WEfRec

Pal Unlo Invc

Recq Maintc TOp

Cor Prob DSt

WEfSto Sales Prod noAvail

Put BuildC CSc CG P

Rate

Pal Sto WH Sto

Stoq Stop OrdProcc Prod Proc

Cor Prob

Ord ProcC Labp

Invq Labc

Cust Ord Scrapq

WH

Pal Moved Rept WEfRep

Repq Nb Scrap

Cor Prob

OrdLTt

Repp WH Rep

WH Del Cust

TrC

WEfDel Complain

Legend: WH Pick

RP – right product WEfShip WH Ship Trc

RQ – right quantity Delp

CustSatq

RT – right truck Pickp Pickt WEfPick Delt

CD – correct document Shipt Shipp Kg Avail

OT – on time

ND – no damage

OrdLi Pick Thp War WH Ord Del TrUtp

P – Profit Pickq OrdLi Ship

CG – Cost of Goods Cor Prob ND Prob

Productivity indicators RP Prob

CD OTDelq

Cost indicators RQ OrdFq

Shipq PerfOrdq

OT

Quality indicators RT OTShipq

Time indicators OT

In Figure C.1 we present data just once to simplify the interpretation. It means that if

there is a data used in two or more indicator equations with dierent units, it will appear

just in one rectangle. That is the case, for example, of Ave Inv that is measured in units

for InvUtp (Equation 5.16) and in dollars for Invc (Equation 5.22) and TOp (Equation

30 C. Manual Procedure to determine indicator relationships

5.17).

The violet blocks referring to Unload pallet Pal Unlo, Pallet stored Pal Sto, Pallet

moved Pal Moved, Order lines picked OrdLiPick, Order lines shipped OrdLiShip and

Orders delivered Ord Del means the total of products processed in each activity. For

these data, we distinguish the main data parts to clarify what is being used to calculate the

indicators. For receiving, storage, replenishment and picking there are just two divisions:

correct `Cor' and problem `Prob'. In the case of order lines shipped and orders delivered,

the acronyms mean, respectively: RP, right product; RQ, right quantity; RT, right truck;

ND, no damage; CD, correct documents; OT, on time. Finally, the red rectangular block,

denoting sales (Equation A.49) is calculated by the sum of prot (represented by the red

The colors denote the classication of indicators and data, according to their dimen-

sions. The green gures refer to data and indicators of time, the red ones refers to cost,

orange to productivity, blue to capacity data and violet is related to the product and order

Figure C.1 shows that the majority of indicators are related with at least one other

indicator, forming a big cloud of relationships. The exceptions are equipment downtime

Analyzing the interconnections, it is possible to visualize some groups formed from this

relations. Taking the left side of Figure C.1, we observe that the violet rectangles (e.g., Pal

Unlo) connect essentially indicators of time, quality and productivity. In the right side of

Figure C.1 it is possible to note a distinct group of indicators mainly associated to costs

can be identied.

In order to clarify the indicator relations, in the next section we present initially a

manual procedure to determine a framework where just indicator relations are exhibited.

After the identication of indicator relations in Figure C.1, we use a simple procedure to

To construct a relationship framework, all indicators are listed and their relations are

identied by means of structures like the one presented in Figure C.2. The indicator under

analysis is located in the center and the ones that are related to it are connected by arrows.

The number on the arrows represents the number of data shared by indicators. Taking one

example of the four demonstrated in Figure C.2, shipping quality Shipq shares one data

with Shipt , Shipp , Thp and two data with OrdFq and OTShipq .

cedure

After the construction of this structure for all indicators, the framework is produced con-

necting indicators with dierent lines depending on the number of data shared. The result

C.2. The indicator relationships schema for the manual procedure 31

Shipt

1

1 1

Shipp Shipq Thp

2 2

OrdFq OTShipq

1 Pickq 1

Pickp Pickt

Recp DSt

Rect 1

1 Put

1

1

1

Pickp CSc

Recq Invq Stop

1 1

1 1

1 1

1 Stop Labc Repp

1

Repq 1 1

Stoq

StockOutq

EqDp WarUtp Invc

OrdProcc

InvUtp Labp Scrapq

Rect Recp

Trc

Pickq

Delp CustSatq

Recq DSt

Pickp Pickt

Invq Put

OrdLTt

Labc PerfOrdq

Stop

Repq

Stoq Shipt

Delt OTDelq

Shipp Thp

Rept Repp

TrUtp

Cost indicators Two Data

Quality indicators One Data

Shipq Time indicators

32 C. Manual Procedure to determine indicator relationships

Looking at Figure C.3, the rst impression could be that the majority of indicators

form a big group of relations. But analyzing Figure C.3 in detail, it is possible to observe

that the indicators are arranged in clusters. The more visible cluster on the right side

of Figure C.3 consists mainly of indicators about delivery process and order quality. The

second group of measures are related to shipping activity, and are located in the bottom

of the gure. The three indicators of picking activity constitute a little group in the center

of the gure. The inbound area, in the left side of the gure, could be viewed as other

important relationship group. However, the relations among inbound indicators do not

seem to be as strong as for the delivery cluster. The last group of measures is located on

the top of the gure, aggregating mainly cost and capacity measures.

It is apparent from Figure C.3 that indicators are rather connected to others by their

processes than by their dimensions. In other words, the indicator relationships seems to

be established per warehouse process, instead of by the dimensions of quality, cost, time,

productivity.

There is two types of lines in Figure C.3: one representing that indicators share one

data and the other one representing two data sharing. We could assume that indicators

with two shared data have a stronger relationship than the others with just one. However,

Figure C.3 shows the main relations, but the procedure performed is not exhaustive.

The analytical model has shown that data are very connected, with some data making part

of more general ones. For example, WH is a sum of all WH Activities(means the sum of

WHRec, WHSto, etc.), as presented in Equation A.22. This situation was not taken into

account in this section. Indeed, Figure C.1 presents WH and WH Activities separately.

To take into account all data associations, next section presents the exhaustive procedure

Appendix D

List of independent input values

α 0.5 mean_Insp 0.5

β_del 0.9 nbMachine 2.0

β_ord 0.55 nb_travel 3.0

β_pick 0.95 NoComplet Ord Ship 17.0

β_rec 0.85 Ord Del OT 1311.0

β_rep 0.8 Ord Ship OT 1334.0

β_ship 0.95 pal_truck 25.0

β_sto 0.85 pallet_area 1.2

BuildC 1988.0 Prob OrdLi Pick 24.0

cap 5000.0 Prob OrdLi Ship 17.0

Cor OrdLi Pick 1367.0 Prob Del 23.0

Cor OrdLi Ship 1334.0 Prob Rep 4.6

Cor Del 1311.0 Prob Sto 2.0

Cor Rep 617.0 Prob Unlo 9.0

Cor Sto 674.0 Prod Ord 18.4

Cor Unlo 691.0 Prod pal 40.0

Cust Ord 1417.0 Prod noAvail 275.0

Cust Complain 18.0 Prod Cost 99.9

ΔT(Insp)2 1.0 Profit 100.0

deprec 1 500.0 Rate 0.1

deprec 2 200.0 Remain_Inv 30500.0

empl Admin 3.0 scrap1 23.0

empl Del 2.0 Scrap_Del1 13.0

empl Pick 4.0 scrap2 5.0

empl Rec 1.0 Scrap_Pick1 4.0

empl Rep 1.0 scrap3 4.0

empl Ship 3.0 scrap4 17.0

empl Sto 1.0 Scrap_Ship1 17.0

EqMaintC 4118.0 scrap5 1.0

error data system1 1.0 scrap6 7.0

error data system2 3.0 Truck Maint C 1165.0

error data system3 1.0 War Cap 5000.0

HAdmindel 63.0 war used area 3800.0

HAdminpick 21.0 War WH 168.0

HAdminrec 21.0 $/hadmin 7.0

HAdminrep 21.0 $/hdel 5.0

HAdminship 21.0 $/hpick 5.0

HAdminsto 21.0 $/hrec 5.0

HEq Stop 14.4 $/hrep 5.0

Inv Cap 1000.0 $/hship 5.0

kg Prod 10.0 $/hsto 5.0

l_used 2.0 $ oil 2.39

Appendix E

Theoretical Framework of indicator relationships

Here we show the theoretical framework of indicator relationships resulted from Jacobian

analysis. To create this schema we perform the same of manual procedure presented in

Appendix C.

Maintc

Scrapq StockOutq

WarUtp Delp

CustSatq

InvUtp TOp

Pickp

Invc PerfOrdq

Pickq

OTDelq

Pickt

CSc

Putt Trc

Stop

OrdLTt

DSt

OrdProcc

Recp

Delt

Invq

Rect

Shipt

Recq Shipp

Thp

EqDp

OTShipq

More than Six Data

Cost indicators

Three up to Six Data

Quality indicators Two Data

Time indicators One Data

Appendix F

Results of Dynamic Factor Analysis application

This appendix reports the initial results obtained with the Dynamic Factor Analysis ap-

plication. The R code and the procedure to perform DFA in R are from Holmes (2015),

The R code is applied for 50 month time series data of the 40 standardized indicators.

The main reason to reduce the dataset to 50 month is because a big dataset does not allow

the convergence of the model. As presented in Chapter 3, Equation 3.9, the objective is to

obtain the Z values, which correspond to the loadings of the PCA method.

Table F.1 demonstrates the DFA results for two dierent R matrix propositions with

the number of trends, m, varying from 1 up to 8. The R matrix measures the covariance

matrix of the observation errors. It can be calculated considering four error conditions:

diagonal and equal, diagonal and unequal, equal variance covariance and unconstrained.

It is shown just two dierent conditions in Table F.1 because are the best results obtained

The logLik (loglikelihood) and the AICc (Akaike Information Criterion with a Correc-

tion for nite sample sizes) are the measures to evaluate the quality of the results. The

lower the logLik and AICc values, better the model. The column K shows the number of

parameters in the model and m represents the number of trends used to represent data.

R m logLik K AICc

diagonal and unequal 1 -2383,03 80,00 4932,81

diagonal and unequal 2 -2096,71 119,00 4446,61

diagonal and unequal 3 -2043,87 157,00 4428,67

diagonal and unequal 4 -1684,87 194,00 3799,65

diagonal and unequal 5 -1542,66 230,00 3605,39

diagonal and unequal 6 -1380,38 265,00 3372,07

diagonal and unequal 7 -1261,68 299,00 3226,89

diagonal and unequal 8 -1200,32 332,00 3197,27

unconstrained 1 70,60 860,00 2878,99

unconstrained 2 112,88 899,00 3043,34

unconstrained 3 166,17 937,00 3196,85

unconstrained 4 205,20 974,00 3390,58

unconstrained 5 236,82 1010,00 3611,29

unconstrained 6 256,13 1045,00 3869,29

unconstrained 7 277,26 1079,00 4136,78

unconstrained 8 295,63 1112,00 4423,39

The bold line in Table F.1 shows the best result for these test: a model with just

EqDp -0,029

Invc -0,049

Invq 0,006

InvUtp 0,243

38 F. Results

Labc of Dynamic

-0,118Factor Analysis application

Labp 0,200

Maintc 0,017

OrdFq 0,188

one trend. Table F.2 shows the loading values obtained and the highlighted cells have

OrdLTt -0,087

|values| > 0.15. It is possible to see that many loadings are really low, resulting that these

OrdProcc -0,117

indicators can not be considered in the model. According to Table F.1, only 11 indicators

OTDelq 0,039

from the initial 40 are included in the aggregated model.

OTShipq 0,145

PerfOrdq 0,028

Indicator Loading Indicator

Pickp Loading

0,106

CSc -0,190 Pickq 0,012

CustSatq -0,029 Pickt -0,022

Delp 0,107 Putt 0,019

Delt -0,109 Recp 0,151

DSt -0,014 Recq 0,022

Rect -0,031

EqDp -0,029

Repp 0,179

Invc -0,049

Repq 0,077

Invq 0,006 Rept -0,002

InvUtp 0,243 Scrapq -0,123

Labc -0,118 Shipp 0,105

Labp 0,200 Shipq 0,209

Maintc 0,017 Shipt 0,004

OrdFq 0,188 StockOutq -0,086

OrdLTt -0,087 Stop 0,145

OrdProcc -0,117 Stoq 0,007

Thp 0,192

OTDelq 0,039

TOp -0,174

OTShipq 0,145

Trc -0,084

PerfOrdq 0,028 TrUtp 0,196

Pickp 0,106 WarUtp 0,163

Pickq 0,012

Pickt -0,022

Table F.2: Loadings for DFA result of m=1 and R= unconstrained.

Putt 0,019

Recpbeen made0,151

Several other tests have but the best results according to the logLik and

Recq 0,022

AICc values are always for m = 1, which exclude a great quantity of indicators from the

Rect -0,031

model. As our objective is to maintain the majority of indicators to evaluate the global

Repp 0,179

performance, we do not use this result in our integrated model.

Repq 0,077

Rept -0,002

Scrapq -0,123

Shipp 0,105

Shipq 0,209

Shipt 0,004

StockOutq -0,086

Stop 0,145

Stoq 0,007

Thp 0,192

TOp -0,174

Trc -0,084

TrUtp 0,196

WarUtp 0,163

Appendix G

Results of Anderson Darling Test

R.

The statistic analysis is performed for each indicator using the software Minitab 16 Each

graphic summarizes the Anderson Darling Test, skewness and kurtosis measurement for all

40 performance indicators. Moreover, the mean and standard deviation are demonstrated

These mean and standard deviation values are used in the optimization model, to

40 G. Results of Anderson Darling Test

Figure G.3: Cost indicator data test. Figure G.4: Time indicator data test.

41

Figure G.5: Quality indicator data test. Figure G.6: Productivity indicator data test.

42 G. Results of Anderson Darling Test

Figure G.7: Quality indicator data test. Figure G.8: Productivity indicator data test.

43

test.

Appendix H

Optimization model

This appendix presents the optimization model coupled with CADES Component Opti-

mizer r .

OBJECTIVE FUNCTION

COMPONENT EQUATIONS

0.5 * Shipq_NORM

0.34 * Scrapq_NORM

46 H. Optimization model

47

Ord_OT_ND_CD = CorDel

CONSTRAINTS

Pal_moved

48 H. Optimization model

TIME INDICATORS

/ (CorUnlo + ProbUnlo)

+ ProbSto)

+ ProbUnlo)

Rep + ProbRep)

+ Prob_OrdLiPick)

/ (Cor_OrdLiShip + Prob_OrdLiShip)

+ ProbDel)

(CorDel + ProbDel)

PRODUCTIVITY INDICATORS

Labp = Product_Ship / WH

49

COST INDICATORS

QUALITY INDICATORS

+ Prob_OrdLiShip))*100

50 H. Optimization model

Appendix I

Mean and standard deviation values of indicators

A complete list of mean and standard deviation values for all indicators are described in

this appendix, Table I.1. The input dataset to obtain this list are the 100 month time

series of each indicator. These values are included as xed variables in the optimization

model.

Standard

Indicator Mean

deviation

CSc 0,36 0,02

CustSatq 99,53 0,39

Delp 3,93 0,46

Delt 0,28 0,03

DSt 0,84 0,10

Invc 221207,91 32507,65

Invq 99,87 0,08

InvUtp 53,51 7,66

Labc 12902,27 810,65

Labp 17,40 1,35

OrdFq 99,27 0,46

OrdLTt 1,25 0,15

OrdProcc 0,90 0,11

OTDelq 99,29 0,41

OTShipq 99,28 0,43

PerfOrdq 99,05 0,50

Pickp 1,98 0,24

Pickt 0,51 0,06

Putt 0,14 0,01

Recp 7,96 0,59

Rect 0,70 0,10

Repp 3,96 0,32

Rept 0,24 0,02

Scrapq 4,27 1,05

Shipp 2,64 0,31

Shipq 99,02 0,52

Shipt 0,38 0,05

Stop 7,96 0,57

Thp 156,64 12,19

TOp 1,33 0,21

Trc 3,26 0,34

TrUtp 84,07 5,43

WarUtp 50,28 2,81

Appendix J

Optimization results

The results of the optimization for the inputs and intermediate outputs are presented,

INPUT RESULTS

Limits in Hours Picking, Shipping and Limits

Time data [unit]

Maximization Minimization Delivery data [unit] Maximization Minimization

β_del 0,34 1,00 Prod noAvail [orders] 3000 0

β_ord 0,30 0,30 No_OT_del [orders] 0 700

β_pick 0,48 1,00 No_OT_ship [orders] 0 700

β_rec 0,53 1,00 No Cust Complain [orders] 3000 0

β_rep 0,41 1,00 NoComplet Ord Ship [orders] 0 0

β_ship 0,48 1,00 Other_Prob_pick [orders] 2 0

β_sto 0,44 1,00 Other_Prob_del [orders] 0 0

Hadmindel [hour] 4,9 1,0 Other_Prob_ship [orders] 0 0

HAdminpick [hour] 1,0 1,0 Cor OrdLi Pick [orders] 3000 660

HAdminrec [hour] 1,0 1,0 Cor OrdLi Ship [orders] 3000 0

HAdminrep [hour] 1,0 1,0 Cor Del [orders] 3000 0

HAdminship [hour] 1,0 1,0 scrap4 [orders] 0 40

HAdminsto [hour] 2,6 141,9 scrap5 [orders] 0 0

scrap6 [orders] 0 0

Limits

Replenishment data [unit]

Maximization Minimization Unloading and Storing data Limits

Cor Rep [pallet] 996 460 [unit] Maximization Minimization

error data system 3 [pallet] 0 0 Cor Sto [pallet] 1000 340

scrap3 [pallet] 0 40 Cor Unlo [pallet] 1000 340

Other_Prob_rep [pallet] 4 0 scrap1 [pallet] 0 15

scrap2 [pallet] 0 18

Limits in $ Other_Prob_sto [pallet] 1,14 0

Cost data

Maximization Minimization Other_Prob_unlo [pallet] 0,5 0

Maintc R$ 1 000,0 R$ 1 000,0 error data system 1 [pallet] 0 4,5

Truck Maint C R$ 50,0 R$ 200 000,0 error data system 2 [pallet] 0 2

Limits

Other data [unit]

Maximization Minimization

War WH [hour] 210 80

Prod Ord [product] 13,3 12,6

war used area [m2] 1000 4000

nb_Travel [travels] 80 300

mean_Insp [h] 0,27 0,5

Cust Ord [orders] 3000 1593

54 J. Optimization results

Results Results

Constraints [unit] Data [unit]

Maximization Minimization Maximization Minimization

CTRL_0 [hour] 45 0,10 aveinv [product] 20000 10000

CTRL_1 [pallet] 0 0 Prob Data [pallet] 0 6,24

CTRL_2 [pallet] 0 0 Cust Complain [orders] 0 700

CTRL_2A [pallet] 0 0 ΔT(Insp) [hour] 10,9 7,7

CTRL_3 [order] 0 893 nb_trucks [trucks] 40 14,39

CTRL_4 [order] 0 0 Prod noAvail [products] 0 0

CTRL_4A [product] 0 0 Ord Del OT [orders] 3000 0

CTRL_5 [order] 0 0 Ord OT, ND, CD [orders] 3000 0

CTRL_6 [order] 0 0 Ord Ship OT [orders] 3000 0

PalProcInv [pallets] 3000 1219

Component Results Prob OrdLi Pick [orders] 2,3 40

Equation Maximization Minimization Prob OrdLi Ship [orders] 0 700

C1 49,40 -184,56 Prob Del [orders] 0 700

C2 24,42 -29,05 Prod Proc [products] 40000 8790

C3 3,42 -50,04 Prob Rep [pallet] 4,3 40

C4 3,49 -193,39 Prob Sto [pallet] 1,2 20

C5 3,52 -282,76 Prob Unlo [pallet] 0,5 20

C6 7,59 0,04 Remain_Inv [products] 0 5605

WarCapUsed 1600 4300

Pal Sto [pallet] 1000 360

Pal Unlo [pallet] 1000 360

Pal Moved [pallet] 1000 500

OrdLi Pick [orders] 3000 700

Ord Ship [orders] 3000 700

Ord Del [orders] 3000 700

Appendix K

Abstracts

Global warehouse management: a methodology to

determine an integrated performance measurement

adopt a large number of indicators, making its management increasingly dicult. Besides

the great quantity of information, it may be hard for managers to assess the interdepen-

dence of indicators with distinct objectives (e.g. the level of a cost indicator shall decrease,

whereas a quality indicator level shall be maximized), making complex the evaluation of

analytical model of performance indicators usually used for warehouse management; (ii)

the denition of indicator relationships analytically and statistically; (iii) the aggregation of

these indicators in an integrated model; (iv) the proposition of a scale to assess the evolution

of the warehouse performance over time according to the integrated model results.

The indicators used to evaluate the warehouse come from the literature and the database is

generated to perform the mathematical tools. The Jacobian matrix is used to dene indi-

cator relationships analytically, and the principal component analysis to achieve indicators'

in six dierent components, which compose the global performance indicator equation by

means of component's weighted average. A scale is developed for the global performance

indicator using an optimization approach to obtain its upper and lower boundaries.

After some tests to verify the usability of the integrated model, we conclude that the

proposed methodology reaches its objective providing a decision support tool for managers

so that they can be more ecient in the global warehouse performance management with-

aggregated indicators, logistics.

Gerenciamento global de armazéns: uma metodologia

para mensurar o desempenho de forma agregada

presas a adotarem um grande número de indicadores de desempenho, o que tem dicultado

cada vez mais o seu gerenciamento. Além do volume de informações, os indicadores normal-

mente possuem interdependências e objetivos distintos, as vezes até opostos (por exemplo,

o indicador de custo deve ser reduzido enquanto o indicador de qualidade deve sempre ser

Dentro deste contexto, esta tese desenvolve uma metodologia para obter uma medida

já utilizados para o gerenciamento do armazém; (ii) a denição das relações entre os indi-

integrado; (iv) a proposição de uma escala para avaliar a evolução do desempenho global

é obtido a partir da média ponderada dos seis componentes. Uma escala é desenvolvida

Depois de testes com o modelo integrado, pôde-se concluir que a metodologia proposta

atingiu seu objetivo ao fornecer uma ferramenta de ajuda à decisão para os gestores, per-

mitindo que eles sejam mais ecazes no gerenciamento global do armazém sem negligenciar

sempenho, indicadores agregados, logística.

Gestion globale des entrepôts logistiques: une

méthodologie pour mesurer la performance de façon

agrégée

RÉSUMÉ: La complexité croissante des opérations dans les entrepôts a conduit les

gestion de plus en plus dicile. De plus, comme ces nombreux indicateurs sont souvent

interdépendants, avec des objectifs diérents, parfois contraires (par exemple, le résultat

d'un indicateur de coût doit diminuer, tandis qu'un indicateur de qualité doit être max-

imisé), il est souvent très dicile pour un manager d'évaluer la performance globale des

Dans ce contexte, cette thèse développe une méthodologie pour atteindre une mesure

utilisés pour la gestion de l'entrepôt; (ii) la dénition de relations entre les indicateurs, de

façon analytique et statistique ; (iii) l'agrégation de ces indicateurs dans un modèle intégré;

La méthodologie est illustrée sur un entrepôt théorique pour démontrer son applica-

littérature, et une base de données est générée pour permettre l'utilisation des outils math-

ématiques. La matrice jacobienne est utilisée pour dénir de façon analytique les relations

entre les indicateurs, et une analyse en composantes principales est faite pour agréger les in-

à partir de la moyenne pondérée de ces six composants. Une échelle est développée pour

Après des testes réalisés avec le modèle intégré, nous concluons que la méthodologie

proposée atteint son objectif en fournissant un outil d'aide à la décision pour les man-

agers an qu'ils puissent être plus ecaces dans la gestion globale de la performance de

l'entrepôt, sans négliger des informations importantes fournis par les indicateurs.

performance, indicateur agrégé, logistique

RÉSUMÉ ÉTENDU

Gestion globale des entrepôts logistiques: une

méthodologie pour mesurer la performance de façon

agrégée

La complexité croissante des opérations dans les entrepôts a conduit les entreprises à utiliser

un grand nombre d'indicateurs de performances, ce qui rend leur gestion de plus en plus

dicile. D'autre part, comme ces nombreux indicateurs sont souvent interdépendants,

peuvent avoir des objectifs diérents, parfois contraires (par exemple, le résultat d'un in-

dicateur de coût doit diminuer, tandis qu'un indicateur de qualité doit être maximisé), il

est souvent très dicile pour un manager d'évaluer la performance globale des systèmes

globale de l'entrepôt d'une manière ecace, en sachant que les personnes ont des limites

de capacité pour traiter une grande quantité d'expressions de performance (Clivillé et al.,

2007). Par conséquent, cette thèse propose un système qui regroupe les indicateurs et

toutes les informations pertinentes. Dans la littérature, plusieurs auteurs ont discuté de

la nécessité d'une mesure globale, mais très peu de travaux ont essayé d'atteindre cet ob-

jectif. Ainsi, les principales lacunes de recherche que cette thèse se propose de combler

sont : dans un ensemble de mesures, si certaines sont bonnes et d'autres sont mauvaises,

comment connaitre la performance globale? (Johnson et al., 2010). Le dé est de con-

cevoir une structure de mesures (par exemple, les regrouper) et en extraire un sens global

situe-t?on?) (Melnyk et al., 2004). De la même façon, Lohman et al. (2004) arme qu'une

question conceptuelle est toujours sans réponse : Quels sont les eets de la combinaison de

plusieurs mesures dans un score global? Au-delà de la critique sur l'utilité d'un indicateur

global et de la possible réticence des gestionnaires pour utiliser les indicateurs agrégés, le

principal dé est de fournir des relations ables entre les indicateurs.

Il est dicile de modéliser les relations entre les indicateurs puisque plusieurs facteurs

inuencent leurs valeurs. De Koster and Balk (2008) illustrent cette situation en armant

que des mesures communes utilisées dans les entrepôts (par exemple les lignes de commande

récupérées par personne et par heure, les taux d'erreur de livraison, les délais de satisfaction

des commandes) ne sont pas mutuellement indépendantes et, en plus, chacune d'elles peut

dépendre de multiples entrées. Le résultat est que les indicateurs ne sont pas seulement

inuencés par un autre indicateur (par exemple, les lignes de commande récupérées par

personne et par heure inuencent la quantité de commandes en retard), mais ils peuvent

K.2 Objectif Général

L'objectif principal de cette thèse est de développer une méthodologie pour l'évaluation de

A partir de l'objectif général présenté, les objectifs spéciques sont proposés comme suit:

données;

• Proposition d'une méthode pour vérier analytiquement les liens entre les indicateurs

de performance;

agrégée.

Les étapes qui ont permis d'atteindre ces objectifs lors de notre travail de thèse, sont

la méthodologie

Pour développer la méthodologie proposée dans cette thèse, plusieurs sujets ont été analysés

Figure K.1: Les sujets étudiés pour développer la méthodologie proposée dans ce travail.

été faite an d'identier les derniers développements et ses possibles lacunes de recherche.

De plus, les résultats obtenus ont mis en évidence les indicateurs de performance les plus

utilisés par les entrepôts logistiques pour évaluer leur performance. En raison des diérents

types d'indicateurs trouvés dans la littérature, certaines classications sont eectuées. Tout

d'abord, nous avons diérencié les indicateurs directs des indicateurs indirects : les indi-

cateurs directs sont dénis par des expressions mathématiques simples tandis que les indi-

cateurs indirects ont besoin d'outils plus sophistiqués de mesure (par exemple l'analyse de

régression, logique oue, DEA, etc.). Après cette étape, les indicateurs directs sont classi-

és selon deux axes (le résultat est présenté dans le Tableau K.1): (i) les lignes du tableau

font la classication selon les dimensions de qualité, coût, temps et productivité; (ii) les

colonnes classient les mesures selon les activités exécutées par l'entrepôt logistique. On

note que certains indicateurs mesurent plusieurs activités en même temps (par exemple,

l'indicateur de délai de satisfaction de la commande, Order lead time); dans ce cas ils

Après cette revue de la littérature, la Figure K.1 montre d'autres sujets étudiés pour

mis de vérier comment les travaux passés ont réussi à proposer une performance globale

sur n'importe quel domaine d'application. En parallèle, les travaux sur le regroupement

d?indicateurs de performance ont montré les outils mathématiques les plus utilisés pour

cette agrégation. Pour appliquer ces outils mathématiques, il est nécessaire de compren-

dre comment ils fonctionnent et quelles sont leurs contraintes. Finalement, pour atteindre

l'objectif de proposer une échelle pour l'indicateur agrégé, des méthodes pour la générer

pour mesurer la performance de l'entrepôt logistique de façon agrégée, ce qui est présenté

Table K.1: Classication des indicateurs directs selon les dimensions et limites des activités.

Receiving Storing Inventory Picking Shipping Delivery

receiving putaway order pick- shipping delivery lead

Time

time time ing time time time

delivery accu-

physical shipping ac-

racy; on-time

storage inventory picking ac- curacy; or-

Quality delivery;

accuracy accuracy; curacy ders shipped

cargo damage

stock-out rate on time

rate

distribution

Cost inventory cost

cost

inventory

receiving picking pro- shipping transport uti-

Productivity space utiliza-

productivity ductivity productivity lization

tion; turnover

Dimensions Process - Transversal Indicators

Inbound Processes Outbound Processes

Dock to stock time Order lead time

Time

Global= Queuing time

Order ll rate, Perfect orders

Quality

Global= Customer satisfaction, Scrap rate

Order processing cost

Cost

Global= Cost as a % of sales

Outbound space utilization

Productivity

Global= Throughput

K.4 Méthodologie pour mesurer la performance de l?entrepôt

logistique de façon agrégée

La méthodologie développée pour atteindre une mesure agrégée de la performance de

l'entrepôt comprend quatre étapes principales (voir Figure K.2): (i) conceptualisation,

La Figure K.2 montre les quatre étapes de la méthodologie avec ses principales sorties.

phase de modélisation mesure de façon théorique les liaisons entre les indicateurs de per-

formance et les agrège en utilisant des outils statistiques. Ces résultats servent de base pour

est faite. De plus, une échelle est créée dans cette phase pour aider à suivre l'évolution de

performance indicators indicator relationships

• Model for indicators

aggregation

Conceptualization Modeling

Implementation

Model Solving

& • Determination of an

• Integrated model

implementation Update Integrated performance

• Model update model

• Scale definition

Figure K.2: Les phases de la méthodologie proposée avec leurs principales étapes. Source:

Adapté de Mitro et al. (1974).

La Figure K.3 détaille les étapes qu'il faut suivre dans chaque phase de la méthodologie

tion des limites d'application de la méthodologie, à savoir, dans quels secteurs de l'entrepôt

la performance sera mesurée et quels seront les indicateurs utilisés pour cela. Cela signie

que, pour appliquer la méthodologie, il est nécessaire de dénir les secteurs où la perfor-

mance sera évaluée et l'ensemble d'indicateurs qui seront utilisés. Ces indicateurs doivent

être connus en termes d?équations, puisque le modèle analytique est formé essentiellement

Figure K.3), il est nécessaire d'acquérir des données à partir des indicateurs. Ces données

sont des séries temporelles obtenues à partir des résultats des indicateurs, qui sont évalués

périodiquement dans l'entrepôt. De cette étape, deux analyses peuvent être menées en

Definition of the scope of performance measurement

Conceptualization

Determination of indicator and data equations

Modeling

Theoretical model of indicator Model for indicators

relationships aggregation

Scale definition

obtenue par l'utilisation d'outils statistiques qui réduisent les dimensions d'un groupe de

variables.

de relations quantitatives entre les indicateurs est construit. Il est appelé modèle de

les valeurs obtenues à partir de ces indicateurs agrégés ne peuvent pas être interprétées

librement, il est nécessaire de créer une échelle pour elles, représentée par l'étape dénition

de l'échelle dans la Figure K.3. Enn, l'étape de mise en oeuvre montre l'utilisation du

modèle pour la gestion périodique d'un entrepôt et la mise à jour dénit le moment où la

cette thèse.

K.5 Application de la méthodologie sur un entrepôt théorique

La méthodologie est illustrée sur un entrepôt théorique pour démontrer son applicabilité.

La Figure K.4 montre l'entrepôt théorique étudié, nommé entrepôt standard. Le nom

standard vient des activités opérationnelles qui sont faites dans l'entrepôt logistique et

qu'on trouve dans la plupart des entrepôts : réception, stockage, réapprovisionnement in-

est eectuée sur les activités opérationnelles, y compris également l'activité de distribution.

La Figure K.4 ne détaille pas seulement les délimitations des activités mais aussi leur em-

placement dans l'entrepôt, et les unités de mesure pour les indicateurs de performance. Les

et une base de données est générée pour permettre l'utilisation des outils mathématiques.

Premièrement, les équations des indicateurs de performance (41 au total) sont dénies.

Comme il y a beaucoup de données utilisées pour mesurer les indicateurs qui ont des liens

entre eux, il fallait aussi élaborer des expressions quantitatives pour les données, an de

trouver toutes les relations possibles entre les indicateurs. Le groupe nal d'expressions,

données.

Pour obtenir les premières idées sur l'agrégation des indicateurs en utilisant des outils

mathématiques, un ensemble de données est nécessaire. Dans un contexte réel, les données

sur les activités de l'entrepôt existent déjà et peuvent être collectées. Cependant, comme

notre entrepôt étudié est théorique, une base de données pour le modèle analytique est

générée, représentant le ux de produits entre les processus. Cette base de données est

utilisée pour calculer les indicateurs de performance de façon mensuelle, ce qui génère une

série temporelle d'indicateurs, qui sont couplés avec les outils mathématiques.

Le premier outil mathématique utilisé est la matrice jacobienne, et les résultats obtenus

performance

La relation quantitative entre les indicateurs est le résultat de diérentes variations et

eets qui se produisent en même temps dans les activités de l'entrepôt. Il est possible

et des données partagées par les indicateurs. L'eet d'enchaînement est l'impact d'un

chaîne d'activités eectuées par l'entrepôt logistique. Dans le cas de l'eet des données

partagées, il détermine que deux indicateurs sont liés par les données qu'ils ont en commun.

L'idée principale de cet eet est que si deux indicateurs ont en commun une ou plusieurs

données, ils ont une sorte de relation, car si une donnée change, les deux indicateurs seront

touchés et vont changer d'une certaine façon. Par exemple, les indicateurs productivité du

travail et taux de perte utilisent la même donnée, quantité de produits traités, dans leur

équation. Si la quantité de produits traités change, les deux indicateurs vont également

En se basant sur le fait que les liaisons entre les indicateurs viennent du partage de

bienne est utilisée pour dénir de façon analytique les relations entre les indicateurs. La

Jacobienne est une matrice des dérivées partielles qui est utilisée pour déterminer la rela-

tion entre les sorties et les entrées (Montgomery and Runger, 2003). En d'autres termes,

la matrice Jacobienne est une dérivée partielle des n sorties (indicateurs de performance)

par rapport aux m entrées (données des indicateurs). Chaque cellule de la matrice donne

la valeur de la dérivée partielle qui peut être interprétée comme la variation de la sortie

r a été utilisé pour mesurer la matrice Jacobienne.

r calcule et donne les résultats numériques

La matrice Jacobienne calculée est d'abord analysée par rapport à ses colonnes. Il y a

principalement deux types d'entrées (les colonnes de la matrice) : celles liées à une seule

sortie et d'autres liées à plusieurs sorties. À titre d'illustration, seulement les parties de la

matrice où les entrées sont liées à deux sorties ou plus sont présentées dans le Tableau K.2,

étant donné que c'est le résultat le plus important pour déterminer les liaisons entre les

indicateurs. Chaque cellule, dans le Tableau K.2, contient les valeurs des dérivées partielles

L'hypothèse présentée, sur l'eet de données partagées, signie que si les dérivées par-

tielles de deux indicateurs sont non nulles pour la même entrée, il existe une relation entre

ces indicateurs. Il est possible de vérier dans le Tableau K.2 que plusieurs indicateurs ont

des entrées communes, ce qui signie une liaison entre eux. Par exemple, les indicateurs

Labc et OrdLTt ont trois entrées communes: βord , emplPick, emplShip, ce qui désigne

une relation. Pour vérier toutes les liaisons qui existent entre les indicateurs de perfor-

INPUTS: 81

INDEPENDENT DATA

OUTPUTS: ALL 41

INDICATORS

JACOBIAN MATRIX:

41 x 81

r : les entrées, les sorties et le résultat de la

matrice Jacobienne.

Table K.2: Vue partielle de la matrice Jacobienne avec les entrées liées à deux sorties ou

plus.

alpha beta_del beta_ord CorDel CorRep CorSto CorUnlo emplPick emplRec emplRep emplShip emplSto

CSc 0,24550 0 0,00000 -0,00038 0 0 0 0,02593 0,02593 0,02593 0,02593 0,02593

CustSatq 0 0 0 0,00101 0 0 0 0 0 0 0 0

Delp 0 0 0 0,00298 0 0 0 0 0 0 0 0

Delt 0 0,25190 0 -0,00021 0 0 0 0 0 0 0 0

DSt 0 0 0 0 0 0 -0,00067 0 0,20400 0 0 0,20400

EqDp 0 0 0 0 0 0 0 0 0 0 0 0

Invc 0 0 0 0 0 199,8 0 0 0 0 0 0

Invq 0 0 0 0 0,00013 0,00013 0,00013 0 0 0 0 0

InvUtp 0 0 0 0 0 0,05000 0 0 0 0 0 0

Labc 9988,0 0 -5292,0 0 0 0 0 1260,0 1260,0 1260,0 1260,0 1260,0

Labp 0 0 0 0 0 0 0 -1,5 -1,5 -1,5 -1,5 -1,5

Maintc 0 0 0 0 0 0 0 0 0 0 0 0

OrdFq 0 0 0 0 0 0 0 0 0 0 0 0

OrdLTt 0 0,25190 0,37780 -0,00101 0 0 0 0,11960 0 0 0,11960 0

OrdProcc 1,4 0 3,7 0 0 0 0 0 0 0 0 0

OTDelq 0 0 0 -0,07367 0 0 0 0 0 0 0 0

Putt 0 0 0 0 0 -0,00036 0 0 0 0 0 0,21120

Recp 0 0 0 0 0 0 0,00595 0 -4,2 0 0 0

Recq 0 0 0 0 0 0 0,00184 0 0 0 0 0

Rect 0 0 0 0 0 0 -0,00033 0 0,20400 0 0 0

Repp 0 0 0 0 0,00595 0 0 0 0 -3,7 0 0

mance, il est nécessaire de comparer toutes les lignes du Tableau K.2, deux par deux. Le

résultat nal de cette analyse est présenté dans le Tableau K.3. Les diérentes couleurs

représentent la quantité des données partagées par deux indicateurs de performance (rouge

pour 1 donnée partagée, bleu pour 2, verte pour 3 ou plus de données partagées). Il est

important de noter que le nombre de données partagées par des indicateurs donnent une

indication sur les relations existant entre eux, mais pas sur l'intensité de ces relations.

La prochaine section applique les outils statistiques pour évaluer les relations entre les

Table K.3: La matrice avec les liaisons entre les indicateurs et la quantité de données partagées.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40

1 CSc 0

2 CustSatq 2 0

3 Delp 3 2 0

4 Delt 3 2 4 0

5 DSt 3 0 1 1 0

6 EqDp 1 0 1 1 1 0

7 Invc 3 0 0 0 0 0 0

8 Invq 0 0 0 0 2 0 2 0

9 InvUtp 0 0 0 0 0 0 4 2 0

10 Labc 15 0 1 1 3 1 0 0 0 0

11 Labp 7 0 1 1 3 1 1 0 0 6 0

12 Maintc 2 0 0 0 0 0 0 0 0 0 0 0

13 OrdFq 0 0 0 0 0 0 0 0 0 0 2 0 0

14 OrdLTt 7 2 4 6 1 1 0 0 0 5 3 0 0 0

15 OrdProcc 6 0 1 1 1 1 0 0 0 5 1 0 0 3 0

16 OTDelq 2 2 2 2 0 0 0 0 0 0 0 0 0 2 0 0

17 OTShipq 0 0 0 0 0 0 0 0 0 0 2 0 2 0 0 0 0

18 PerfOrdq 2 2 2 2 0 0 0 0 0 0 0 0 0 2 0 2 0 0

19 Pickp 2 0 1 1 1 1 2 0 0 2 2 0 0 2 1 0 0 0 0

20 Pickq 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 2 0

21 Pickt 2 0 1 1 1 1 2 0 0 2 2 0 0 4 1 0 0 0 4 2 0

22 Putt 2 0 1 1 4 1 2 2 2 2 2 0 0 1 1 0 0 0 1 0 1 0

23 Recp 2 0 1 1 4 1 0 2 0 2 2 0 0 1 1 0 0 0 1 0 1 1 0

24 Recq 0 0 0 0 2 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0

25 Rect 2 0 1 1 8 1 0 2 0 2 2 0 0 1 1 0 0 0 1 0 1 1 4 2 0

26 Repp 2 0 1 1 1 1 0 2 0 2 2 0 0 1 1 0 0 0 1 0 1 1 1 0 1 0

27 Repq 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0

28 Rept 2 0 1 1 1 1 0 2 0 2 2 0 0 1 1 0 0 0 1 0 1 1 1 0 1 4 2 0

29 Scrapq 1 0 0 0 0 0 2 0 1 0 4 0 2 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0

30 Shipp 2 0 1 1 1 1 0 0 0 2 4 0 2 2 1 0 2 0 1 0 1 1 1 0 1 1 0 1 2 0

31 Shipq 0 0 0 0 0 0 0 0 0 0 2 0 2 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 2 2 0

32 Shipt 2 0 1 1 1 1 0 0 0 2 4 0 2 5 1 0 2 0 1 0 1 1 1 0 1 1 0 1 2 4 2 0

33 StockOutq 1 0 0 0 0 0 5 0 0 0 1 0 0 0 0 0 0 0 2 2 2 0 0 0 0 0 0 0 1 0 0 0 0

34 Stop 2 0 1 1 2 1 2 2 2 2 2 0 0 1 1 0 0 0 1 0 1 4 1 0 1 1 0 1 0 1 0 1 0 0

35 Stoq 0 0 0 0 0 0 2 2 2 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 2 0

36 Thp 2 0 1 1 1 1 1 0 0 1 5 0 2 1 1 0 2 0 1 0 1 1 1 0 1 1 0 1 4 3 2 3 1 1 0 0

37 TOp 4 2 2 2 0 0 5 2 4 0 1 0 0 2 0 2 0 2 0 0 0 2 0 0 0 0 0 0 2 0 0 0 1 2 2 1 0

38 Trc 4 2 4 4 1 1 0 0 0 2 1 0 0 4 2 2 0 2 1 0 1 1 1 0 1 1 0 1 0 1 0 1 0 1 0 1 2 0

39 TrUtp 4 2 2 2 0 0 1 0 0 0 1 0 0 2 0 2 0 2 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 1 4 3 0

40 WarUtp 0 0 0 0 0 0 4 2 4 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 1 0 0 0 0 2 2 0 4 0 0 0

K.5.2 Agrégation d'indicateurs de performance par des outils statis-

tiques - premiers résultats

Parmi les outils statistiques pouvant être utilisés pour agréger les indicateurs, le PCA

(Analyse de Composantes Principales) a été choisi dans ce travail parce qu'il n'a pas de

contraintes sur l'utilisation de séries temporelles comme données d'entrée, et aussi pour sa

facilité d'application. Vu que le PCA agrège les indicateurs à partir de leurs corrélations

statistiques, la mesure de la matrice des corrélations entre les indicateurs est présentée

Les numéros à l'intérieur du Tableau K.4 sont les coecients de corrélation, nommés r

de Pearson (ou simplement r ). Toutes les cellules sélectionnées présentent une corrélation

signicative, avec la valeur de p < 0,01. Les cellules bleues montrent des corrélations

moyennes, avec des valeurs absolues des coecients de corrélation comprises entre 0,4 et

0,59; et les cellules roses montrent les corrélations élevées, avec des valeurs absolues des

Il est possible de vérier que certains indicateurs dans le Tableau K.4 n'ont que des

corrélations faibles, ou juste quelques corrélations moyennes. Par exemple, EqDp , Invq et

Maintc ne disposent pas de corrélations supérieures à 0,4 (|r| ≥ 0, 4). Ce type de résultat

montre que ces indicateurs peuvent avoir des problèmes au moment d'être incorporés dans

les résultats du PCA, puisque les composants sont constitués sur les corrélations entre les

variables.

des 40 indicateurs de performance. Avant d'insérer les données dans l'outil PCA, ces

données ont été standardisées à cause de la sensibilité du modèle aux grandes variations

que les données peuvent présenter. L'objectif de ce PCA est de vérier le comportement

des indicateurs dans les situations d'agrégation, ce qui fournit plus d'éléments pour dénir

le groupe nal d'indicateurs qui fera partie du modèle agrégé. Le résultat de cette première

La Figure K.6 et les suivantes, qui montrent le résultat d'un PCA, sont divisées en

cumulative de la variance pour les composantes principales (en bas de la gure); un tableau

plot dans la partie droite de la gure. Chacune de ces trois parties est expliquée dans ce

qui suit.

yser. Dans un premier temps, l'écart type de chaque composante principale, lorsqu'il est

supérieur à un, est déni comme l'un des critères pour choisir les composantes à retenir.

A titre d'exemple, dans la Figure K.6 il y a 10 composants (PC1 jusqu'à PC10) avec un

écart type supérieur à un, ce qui indique que ces dix éléments doivent être considérés dans

variance de données, tandis que la proportion cumulée (troisième ligne) présente la somme

des écarts types pour toutes les composantes. À partir du choix des dix composants, la

proportion cumulée est de 86,55 %, ce qui signie que les dix éléments expliquent 86,55 %

Table K.4: Matrice des corrélations statistiques.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40

1 CSc 1

2 CustSatq -0,1 1

3 Delp -0,6 0 1

4 Delt 0,65 -0 -0,98 1

5 DSt 0,02 -0,1 -0,16 0,13 1

6 EqDp -0,1 0,1 0,07 -0,1 -0,2 1

7 Invc 0,1 -0,2 -0,04 0,03 0,16 -0 1

8 Invq -0,1 -0,1 0,07 -0,1 -0,1 -0 0,09 1

9 InvUtp 0,11 -0,2 -0,02 0,02 0,16 -0 0,97 0,1 1

10 Labc 0,55 -0,2 -0,52 0,52 0,1 -0 0,14 -0 0,12 1

11 Labp -0,96 0,1 0,67 -0,7 -0,1 0,1 -0,13 0,1 -0,1 -0,7 1

12 Maintc -0 -0,1 0,11 -0,1 -0 -0 0,03 -0 0,04 0,21 0,07 1

13 OrdFq 0,04 -0,2 0 0,01 0 -0 0,11 0,1 0,1 0,06 -0 -0 1

14 OrdLTt 0,65 -0 -0,98 1,0 0,13 -0 0,03 -0 0,02 0,52 -0,7 -0,1 0 1

15 OrdProcc 0,60 0 -0,97 0,99 0,14 -0 0,04 -0 0,02 0,43 -0,6 -0,1 0 0,99 1

16 OTDelq -0 0,6 -0,15 0,12 0,05 -0 -0,05 0 -0 -0 -0 -0 -0,2 0,12 0,12 1

17 OTShipq 0,07 -0,2 -0,03 0,04 -0 -0 0,03 0 0,03 0,06 -0 -0 0,9 0,04 0,05 -0,2 1

18 PerfOrdq -0 0,7 -0,03 0,03 -0 0,1 -0,03 -0 -0 -0 -0 -0 -0,1 0,03 0,02 0,9 -0,2 1

19 Pickp -0,6 0 1 -0,97 -0,2 0,1 -0,05 0,1 -0 -0,5 0,66 0,11 0 -0,97 -0,97 -0,1 -0 -0 1

20 Pickq -0,1 0,1 0,07 -0,1 -0 0,1 -0,18 -0 -0,1 -0,1 0,11 -0,1 -0,2 -0,1 -0,1 0,1 -0,1 0,1 0,04 1

21 Pickt 0,63 -0 -0,97 1,00 0,13 -0 0,03 -0 0,02 0,51 -0,7 -0,1 0 1,0 0,99 0,1 0 0 -0,98 -0,1 1

22 Putt 0,38 -0,1 -0,32 0,33 -0,2 -0 -0,12 -0 -0,1 0,74 -0,5 0,12 0 0,33 0,24 -0,1 0,1 -0,1 -0,3 -0 0,31 1

23 Recp -0,39 0,1 0,34 -0,3 0,15 0,2 0,11 0,1 0,14 -0,76 0,51 -0,1 0 -0,3 -0,3 0,1 -0,1 0,1 0,33 0,01 -0,3 -1 1

24 Recq 0,08 0,1 -0,21 0,19 0,1 -0 0,01 0,2 0,04 0 -0,1 -0,1 -0,1 0,19 0,22 0,1 -0,1 0,1 -0,2 0,09 0,21 -0,1 0,04 1

25 Rect -0 -0,1 -0,12 0,1 1,00 -0 0,17 -0 0,17 0,02 -0 -0 0 0,1 0,11 0,1 -0 -0 -0,1 -0 0,1 -0,3 0,24 0,1 1

26 Repp -0,95 0,1 0,69 -0,7 -0,1 0,1 -0,14 0,1 -0,1 -0,69 0,98 0,07 -0 -0,7 -0,7 0 -0,1 0 0,68 0,08 -0,7 -0,5 0,5 -0,1 -0 1

27 Repq -0,1 -0 0,06 -0 0,03 -0 0,18 0,1 0,19 -0,1 0,11 0,07 0,1 -0 -0 0 0,1 -0 0,06 -0,1 -0 -0,1 0,13 0,1 0,04 0,04 1

28 Rept 0,96 -0,1 -0,7 0,73 0,06 -0 0,13 -0 0,12 0,69 -0,98 -0,1 0 0,73 0,68 -0 0,1 -0 -0,7 -0,1 0,72 0,47 -0,5 0,1 0,01 -0,99 -0 1

29 Scrapq 0,08 -0,3 0,03 -0 -0 0 -0,02 0,1 -0 0,12 -0,1 0,09 -0,2 -0 -0 -0,4 -0,3 -0,4 0,04 -0,3 -0 0,12 -0,1 -0,4 -0 -0 -0,41 0,03 1

30 Shipp -0,6 0 1,0 -0,98 -0,2 0,1 -0,05 0,1 -0 -0,5 0,67 0,11 -0 -0,98 -0,97 -0,1 -0,1 -0 1,0 0,07 -0,97 -0,3 0,34 -0,2 -0,1 0,69 0,06 -0,7 0,04 1

31 Shipq 0,05 -0,1 0,03 -0 -0 -0 0,07 -0 0,08 0,06 -0 0 0,8 -0 -0 -0,1 0,8 -0 0,05 -0,2 -0 0,02 -0 -0,2 -0 -0 0,03 0,02 -0,3 0,01 1

32 Shipt 0,65 -0 -0,98 1,00 0,13 -0 0,03 -0 0,02 0,52 -0,7 -0,1 0 1,00 0,99 0,1 0,1 0 -0,97 -0,1 1 0,33 -0,3 0,2 0,09 -0,7 -0 0,73 -0 -0,98 -0 1

33 StockOutq 0,06 -0,1 -0,04 0,03 0,08 -0 0,4 0 0,19 0,11 -0,1 -0,1 0,1 0,03 0,02 -0,1 0,1 -0 -0 -0,5 0,01 0,04 -0,1 -0,1 0,07 -0,1 0,09 0,08 0,05 -0 0,07 0,03 1

34 Stop -0,4 0,1 0,32 -0,3 0,16 0,2 0,14 0,1 0,16 -0,7 0,48 -0,1 -0 -0,3 -0,2 0,1 -0,1 0,1 0,31 0,01 -0,3 -0,99 0,99 0,1 0,25 0,48 0,14 -0,47 -0,1 0,32 -0 -0,3 -0 1

35 Stoq -0 0,2 0,06 -0,1 -0,2 -0 -0,15 0,2 -0,1 -0,2 0,04 -0,2 0,2 -0,1 -0 0,1 0,1 0 0,06 0,01 -0 -0,1 0,18 0,1 -0,1 0,05 0,01 -0,04 -0,46 0,06 0,2 -0 -0,1 0,14 1

36 Thp -0,96 0,1 0,67 -0,7 -0,1 0,1 -0,13 0,1 -0,1 -0,69 1 0,07 -0 -0,7 -0,6 -0 -0 -0 0,66 0,11 -0,7 -0,5 0,51 -0,1 -0 0,98 0,11 -0,98 -0,1 0,67 -0 -0,7 -0,1 0,48 0,04 1

37 TOp -0,4 0,1 0,17 -0,2 -0,1 0 -0,88 -0 -0,91 -0,1 0,38 0,08 -0,1 -0,2 -0,2 0 -0,1 0 0,17 0,13 -0,2 0,14 -0,1 -0,1 -0,1 0,39 -0,2 -0,38 0,04 0,17 -0,1 -0,2 -0,2 -0,2 0,07 0,38 1

38 Trc 0,59 0 -0,96 0,98 0,14 -0 0,01 -0 -0 0,41 -0,6 -0,1 -0 0,98 0,98 0,1 0 0 -0,96 -0,1 0,98 0,24 -0,3 0,2 0,11 -0,6 -0 0,66 -0 -0,97 -0 0,98 0 -0,2 -0 -0,6 -0,2 1

39 TrUtp -0,5 0,2 0,4 -0,4 -0,1 0,2 -0,08 0,2 -0,1 -0,96 0,64 -0,2 -0 -0,4 -0,3 0,1 -0 0,1 0,37 0,07 -0,4 -0,8 0,77 0,1 0,02 0,62 0,09 -0,62 -0,1 0,4 -0,1 -0,4 -0,1 0,75 0,18 0,64 0,04 -0,3 1

40 WarUtp -0,1 -0,1 0,01 -0,1 0,12 -0 0,6 0,1 0,61 0,14 0,01 0,01 0,1 -0,1 -0,1 -0,1 0 -0,2 0,01 -0,1 -0 0,06 -0,1 0,1 0,11 0,03 0,04 -0,04 0,04 0,01 0,12 -0,1 0,2 -0,1 -0,1 0,01 -0,5 -0,1 -0 1

Le tableau d'indicateurs versus composantes montre, dans les cellules, les charges aij

qui donnent le poids de chaque indicateur dans la composante respective. Les cellules

en surbrillance sont celles qui vérient |charge| ≥ 0, 3, désignant les indicateurs pris en